Exemple #1
0
    def test_trajectory(self):
        q_ref = np.ones(5)
        traj_euclidian = tsid.TrajectoryEuclidianConstant("traj_eucl", q_ref)

        if sys.version_info >= (3, 2):
            with self.assertWarns(UserWarning):
                traj_euclidian.computeNext().pos()
            with self.assertWarns(UserWarning):
                traj_euclidian.computeNext().vel()
            with self.assertWarns(UserWarning):
                traj_euclidian.computeNext().acc()
        else:
            with warnings.catch_warnings(record=True) as w:
                self.assertEqual(len(w), 0)

                traj_euclidian.computeNext().pos()
                self.assertEqual(len(w), 1)
                self.assertEqual(w[-1].category, UserWarning)

                traj_euclidian.computeNext().vel()
                self.assertEqual(len(w), 2)
                self.assertEqual(w[-1].category, UserWarning)

                traj_euclidian.computeNext().acc()
                self.assertEqual(len(w), 3)
                self.assertEqual(w[-1].category, UserWarning)
    def control(self, qmes12, vmes12, t):

        if (t > 3.0):
            self.com_new_ref = np.sin(self.omega * t) + self.com_ref

            self.trajCom = tsid.TrajectoryEuclidianConstant(
                "traj_com", self.com_new_ref)

            self.sampleCom = self.trajCom.computeNext()
            self.comTask.setReference(self.sampleCom)

        # Resolution of the HQP problem
        HQPData = self.invdyn.computeProblemData(t, self.qdes, self.vdes)
        self.sol = self.solver.solve(HQPData)

        # Torques, accelerations, velocities and configuration computation
        tau_ff = self.invdyn.getActuatorForces(self.sol)
        self.ades = self.invdyn.getAccelerations(self.sol)
        self.vdes += self.ades * dt
        self.qdes = pin.integrate(self.model, self.qdes, self.vdes * dt)

        # Torque PD controller
        P = 50.0
        D = 0.2
        torques12 = P * (self.qdes[7:] - qmes12[7:]) + D * (
            self.vdes[6:] - vmes12[6:]) + tau_ff

        torques8 = np.concatenate(
            (torques12[1:3], torques12[4:6], torques12[7:9], torques12[10:12]))

        # Saturation to limit the maximal torque
        t_max = 2.5
        tau = np.maximum(np.minimum(torques8, t_max * np.ones((8, 1))),
                         -t_max * np.ones((8, 1)))

        self.error = self.error or (self.sol.status != 0) or (
            qmes12[8] < -np.pi / 2) or (qmes12[11] < -np.pi / 2) or (
                qmes12[14] < -np.pi / 2) or (qmes12[17] < -np.pi / 2) or (
                    qmes12[8] > np.pi / 2) or (qmes12[11] > np.pi / 2) or (
                        qmes12[14] > np.pi / 2) or (qmes12[17] > np.pi / 2)

        return tau.flatten()
Exemple #3
0
    def log(self, t, solo, k_simu, sequencer, interface):
        """ To log various quantities
        Not used anymore since logging is done by the Logger object
        """

        self.h_ref_feet[k_simu] = interface.mean_feet_z

        # Log pos, vel, acc of the flying foot
        for i_foot in range(4):
            self.f_pos_ref[i_foot, k_simu:(k_simu+1), :] = self.sampleFeet[i_foot].pos()[0:3].transpose()
            self.f_vel_ref[i_foot, k_simu:(k_simu+1), :] = self.sampleFeet[i_foot].vel()[0:3].transpose()
            self.f_acc_ref[i_foot, k_simu:(k_simu+1), :] = self.sampleFeet[i_foot].acc()[0:3].transpose()

            pos = self.robot.framePosition(self.invdyn.data(), self.ID_feet[i_foot])
            vel = self.robot.frameVelocityWorldOriented(
                self.invdyn.data(), self.ID_feet[i_foot])
            acc = self.robot.frameAccelerationWorldOriented(
                self.invdyn.data(), self.ID_feet[i_foot])
            self.f_pos[i_foot, k_simu:(k_simu+1), :] = interface.o_feet[:,
                                                                            i_foot]  # pos.translation[0:3].transpose()
            self.f_vel[i_foot, k_simu:(k_simu+1), :] = vel.vector[0:3].transpose()
            self.f_acc[i_foot, k_simu:(k_simu+1), :] = acc.vector[0:3].transpose()

        c_f = np.zeros((3, 4))
        for i, j in enumerate(np.where(sequencer.S[0, :] == 1)[0]):
            c_f[:, j:(j+1)] = self.fc[(3*i):(3*(i+1))]
        self.c_forces[:, k_simu, :] = c_f.transpose()  #  self.fc.reshape((4, 3))

        # Log position of the base
        pos_trunk = self.robot.framePosition(self.invdyn.data(), self.ID_base)
        self.b_pos[k_simu:(k_simu+1), 0:3] = pos_trunk.translation[0:3].transpose()
        vel_trunk = self.robot.frameVelocityWorldOriented(self.invdyn.data(), self.ID_base)
        self.b_vel[k_simu:(k_simu+1), 0:3] = vel_trunk.vector[0:3].transpose()

        # Log position and reference of the CoM
        com_ref = self.robot.com(self.invdyn.data())
        self.trajCom = tsid.TrajectoryEuclidianConstant("traj_com", com_ref)
        sample_com = self.trajCom.computeNext()

        self.com_pos_ref[k_simu:(k_simu+1), 0:3] = sample_com.pos().transpose()
        self.com_pos[k_simu:(k_simu+1), 0:3] = self.robot.com(self.invdyn.data()).transpose()
Exemple #4
0
    def __init__(self, conf, viewer=True):
        self.conf = conf
        vector = se3.StdVec_StdString()
        vector.extend(item for item in conf.path)
        self.robot = tsid.RobotWrapper(conf.urdf, vector, False)
        robot = self.robot
        self.model = model = robot.model()
        try:
            #            q = se3.getNeutralConfiguration(model, conf.srdf, False)
            se3.loadReferenceConfigurations(model, conf.srdf, False)
            q = model.referenceConfigurations['default']
#        q = model.referenceConfigurations["half_sitting"]
        except:
            q = conf.q0
#            q = np.matrix(np.zeros(robot.nv)).T
        v = np.matrix(np.zeros(robot.nv)).T

        assert model.existFrame(conf.ee_frame_name)

        formulation = tsid.InverseDynamicsFormulationAccForce(
            "tsid", robot, False)
        formulation.computeProblemData(0.0, q, v)

        postureTask = tsid.TaskJointPosture("task-posture", robot)
        postureTask.setKp(conf.kp_posture * matlib.ones(robot.nv).T)
        postureTask.setKd(2.0 * np.sqrt(conf.kp_posture) *
                          matlib.ones(robot.nv).T)
        formulation.addMotionTask(postureTask, conf.w_posture, 1, 0.0)

        self.eeTask = tsid.TaskSE3Equality("task-ee", self.robot,
                                           self.conf.ee_frame_name)
        self.eeTask.setKp(self.conf.kp_ee * np.matrix(np.ones(6)).T)
        self.eeTask.setKd(2.0 * np.sqrt(self.conf.kp_ee) *
                          np.matrix(np.ones(6)).T)
        self.eeTask.setMask(conf.ee_task_mask)
        self.eeTask.useLocalFrame(False)
        self.EE = model.getFrameId(conf.ee_frame_name)
        H_ee_ref = self.robot.framePosition(formulation.data(), self.EE)
        self.trajEE = tsid.TrajectorySE3Constant("traj-ee", H_ee_ref)
        formulation.addMotionTask(self.eeTask, conf.w_ee, 1, 0.0)

        self.tau_max = conf.tau_max_scaling * model.effortLimit
        self.tau_min = -self.tau_max
        actuationBoundsTask = tsid.TaskActuationBounds("task-actuation-bounds",
                                                       robot)
        actuationBoundsTask.setBounds(self.tau_min, self.tau_max)
        if (conf.w_torque_bounds > 0.0):
            formulation.addActuationTask(actuationBoundsTask,
                                         conf.w_torque_bounds, 0, 0.0)

        jointBoundsTask = tsid.TaskJointBounds("task-joint-bounds", robot,
                                               conf.dt)
        self.v_max = conf.v_max_scaling * model.velocityLimit
        self.v_min = -self.v_max
        jointBoundsTask.setVelocityBounds(self.v_min, self.v_max)
        if (conf.w_joint_bounds > 0.0):
            formulation.addMotionTask(jointBoundsTask, conf.w_joint_bounds, 0,
                                      0.0)

        trajPosture = tsid.TrajectoryEuclidianConstant("traj_joint", q)
        postureTask.setReference(trajPosture.computeNext())

        solver = tsid.SolverHQuadProgFast("qp solver")
        solver.resize(formulation.nVar, formulation.nEq, formulation.nIn)

        self.trajPosture = trajPosture
        self.postureTask = postureTask
        self.actuationBoundsTask = actuationBoundsTask
        self.jointBoundsTask = jointBoundsTask
        self.formulation = formulation
        self.solver = solver
        self.q = q
        self.v = v

        # for gepetto viewer
        if (viewer):
            self.robot_display = se3.RobotWrapper.BuildFromURDF(
                conf.urdf, [
                    conf.path,
                ])
            l = subprocess.getstatusoutput(
                "ps aux |grep 'gepetto-gui'|grep -v 'grep'|wc -l")
            if int(l[1]) == 0:
                os.system('gepetto-gui &')
            time.sleep(1)
            gepetto.corbaserver.Client()
            self.robot_display.initViewer(loadModel=True)
            self.robot_display.displayCollisions(False)
            self.robot_display.displayVisuals(True)
            self.robot_display.display(q)
            self.gui = self.robot_display.viewer.gui
            self.gui.setCameraTransform(0, conf.CAMERA_TRANSFORM)
import pinocchio as se3
import tsid
import numpy as np

print ""
print "Test Trajectory Euclidian"
print ""

tol = 1e-5
n = 5
q_ref = np.matrix(np.ones(n)).transpose()
zero = np.matrix(np.zeros(n)).transpose()

traj_euclidian = tsid.TrajectoryEuclidianConstant("traj_eucl", q_ref)
assert traj_euclidian.has_trajectory_ended()
assert np.linalg.norm(traj_euclidian.computeNext().pos() - q_ref, 2) < tol
assert np.linalg.norm(traj_euclidian.getSample(0.0).pos() - q_ref, 2) < tol

traj_sample = tsid.TrajectorySample(n)
traj_euclidian.getLastSample(traj_sample)
assert np.linalg.norm(traj_sample.pos() - q_ref, 2) < tol
assert np.linalg.norm(traj_sample.vel() - zero, 2) < tol
assert np.linalg.norm(traj_sample.acc() - zero, 2) < tol

print ""
print "Test Trajectory SE3"
print ""

M_ref = se3.SE3.Identity()
M_vec = np.matrix(np.zeros(12)).transpose()
M_vec[0:3] = M_ref.translation
Exemple #6
0
    def __init__(self, N_simulation, k_mpc, n_periods):

        self.q_ref = np.array([[0.0, 0.0, 0.2027682, 0.0, 0.0, 0.0, 1.0,
                                0.0, 0.8, -1.6, 0, 0.8, -1.6,
                                0, -0.8, 1.6, 0, -0.8, 1.6]]).transpose()

        self.qtsid = self.q_ref.copy()
        self.vtsid = np.zeros((18, 1))
        self.ades = np.zeros((18, 1))

        self.error = False
        self.verbose = True

        # List with the names of all feet frames
        self.foot_frames = ['FL_FOOT', 'FR_FOOT', 'HL_FOOT', 'HR_FOOT']

        # Constraining the contacts
        mu = 0.9 				# friction coefficient
        fMin = 1.0				# minimum normal force
        fMax = 25.0  			# maximum normal force
        contactNormal = np.matrix([0., 0., 1.]).T  # direction of the normal to the contact surface

        # Coefficients of the posture task
        kp_posture = 10.0		# proportionnal gain of the posture task
        w_posture = 1.0         # weight of the posture task

        # Coefficients of the contact tasks
        kp_contact = 100.0         # proportionnal gain for the contacts
        self.w_forceRef = 50.0  # weight of the forces regularization
        self.w_reg_f = 50.0

        # Coefficients of the foot tracking task
        kp_foot = 100.0               # proportionnal gain for the tracking task
        self.w_foot = 500.0       # weight of the tracking task

        # Arrays to store logs
        k_max_loop = N_simulation
        self.f_pos = np.zeros((4, k_max_loop, 3))
        self.f_vel = np.zeros((4, k_max_loop, 3))
        self.f_acc = np.zeros((4, k_max_loop, 3))
        self.f_pos_ref = np.zeros((4, k_max_loop, 3))
        self.f_vel_ref = np.zeros((4, k_max_loop, 3))
        self.f_acc_ref = np.zeros((4, k_max_loop, 3))
        self.b_pos = np.zeros((k_max_loop, 6))
        self.b_vel = np.zeros((k_max_loop, 6))
        self.com_pos = np.zeros((k_max_loop, 3))
        self.com_pos_ref = np.zeros((k_max_loop, 3))
        self.c_forces = np.zeros((4, k_max_loop, 3))
        self.h_ref_feet = np.zeros((k_max_loop, ))
        self.goals = np.zeros((3, 4))
        self.vgoals = np.zeros((3, 4))
        self.agoals = np.zeros((3, 4))
        self.mgoals = np.zeros((6, 4))

        # Position of the shoulders in local frame
        self.shoulders = np.array([[0.19, 0.19, -0.19, -0.19], [0.15005, -0.15005, 0.15005, -0.15005]])
        self.footsteps = self.shoulders.copy()
        self.memory_contacts = self.shoulders.copy()

        # Foot trajectory generator
        max_height_feet = 0.05
        t_lock_before_touchdown = 0.05
        self.ftgs = [ftg.Foot_trajectory_generator(max_height_feet, t_lock_before_touchdown) for i in range(4)]

        # Which pair of feet is active (0 for [1, 2] and 1 for [0, 3])
        self.pair = -1

        # Number of TSID steps for 1 step of the MPC
        self.k_mpc = k_mpc

        # For update_feet_tasks function
        self.dt = 0.001  #  [s], time step
        self.t1 = 0.14  # [s], duration of swing phase

        # Rotation matrix
        self.R = np.eye(3)

        # Feedforward torques
        self.tau_ff = np.zeros((12, 1))

        # Torques sent to the robot
        self.torques12 = np.zeros((12, 1))
        self.tau = np.zeros((12, ))

        self.ID_base = None  # ID of base link
        self.ID_feet = [None] * 4  # ID of feet links

        # Footstep planner object
        # self.fstep_planner = FootstepPlanner.FootstepPlanner(0.001, 32)
        self.vu_m = np.zeros((6, 1))
        self.t_stance = 0.16
        self.T_gait = 0.32
        self.n_periods = n_periods
        self.h_ref = 0.235 - 0.01205385
        self.t_swing = np.zeros((4, ))  # Total duration of current swing phase for each foot

        self.contacts_order = [0, 1, 2, 3]

        # Parameter to enable/disable hybrid control
        self.enable_hybrid_control = False

        # Time since the start of the simulation
        self.t = 0.0

        ########################################################################
        #             Definition of the Model and TSID problem                 #
        ########################################################################

        # Set the paths where the urdf and srdf file of the robot are registered
        modelPath = "/opt/openrobots/share/example-robot-data/robots"
        urdf = modelPath + "/solo_description/robots/solo12.urdf"
        srdf = modelPath + "/solo_description/srdf/solo.srdf"
        vector = pin.StdVec_StdString()
        vector.extend(item for item in modelPath)

        # Create the robot wrapper from the urdf model (which has no free flyer) and add a free flyer
        self.robot = tsid.RobotWrapper(urdf, vector, pin.JointModelFreeFlyer(), False)
        self.model = self.robot.model()

        # Creation of the Invverse Dynamics HQP problem using the robot
        # accelerations (base + joints) and the contact forces
        self.invdyn = tsid.InverseDynamicsFormulationAccForce("tsid", self.robot, False)

        # Compute the problem data with a solver based on EiQuadProg
        t = 0.0
        self.invdyn.computeProblemData(t, self.qtsid, self.vtsid)

        # Saving IDs for later
        self.ID_base = self.model.getFrameId("base_link")
        for i, name in enumerate(self.foot_frames):
            self.ID_feet[i] = self.model.getFrameId(name)

        # Store a frame object to avoid creating one each time
        self.pos_foot = self.robot.framePosition(self.invdyn.data(), self.ID_feet[0])

        #####################
        # LEGS POSTURE TASK #
        #####################

        # Task definition (creating the task object)
        self.postureTask = tsid.TaskJointPosture("task-posture", self.robot)
        self.postureTask.setKp(kp_posture * matlib.ones(self.robot.nv-6).T)  # Proportional gain
        self.postureTask.setKd(2.0 * np.sqrt(kp_posture) * matlib.ones(self.robot.nv-6).T)  # Derivative gain

        # Add the task to the HQP with weight = w_posture, priority level = 1 (not real constraint)
        # and a transition duration = 0.0
        self.invdyn.addMotionTask(self.postureTask, w_posture, 1, 0.0)

        # TSID Trajectory (creating the trajectory object and linking it to the task)
        pin.loadReferenceConfigurations(self.model, srdf, False)
        self.trajPosture = tsid.TrajectoryEuclidianConstant("traj_joint", self.q_ref[7:])
        self.samplePosture = self.trajPosture.computeNext()
        self.postureTask.setReference(self.samplePosture)

        ############
        # CONTACTS #
        ############

        self.contacts = 4*[None]  # List to store the rigid contact tasks

        for i, name in enumerate(self.foot_frames):

            # Contact definition (creating the contact object)
            self.contacts[i] = tsid.ContactPoint(name, self.robot, name, contactNormal, mu, fMin, fMax)
            self.contacts[i].setKp((kp_contact * matlib.ones(3).T))
            self.contacts[i].setKd((2.0 * np.sqrt(kp_contact) * matlib.ones(3).T))
            self.contacts[i].useLocalFrame(False)

            # Set the contact reference position
            H_ref = self.robot.framePosition(self.invdyn.data(), self.ID_feet[i])
            H_ref.translation = np.matrix(
                [H_ref.translation[0, 0],
                 H_ref.translation[1, 0],
                 0.0]).T
            self.contacts[i].setReference(H_ref)

            # Regularization weight for the force tracking subtask
            self.contacts[i].setRegularizationTaskWeightVector(
                np.matrix([self.w_reg_f, self.w_reg_f, self.w_reg_f]).T)

            # Adding the rigid contact after the reference contact force has been set
            self.invdyn.addRigidContact(self.contacts[i], self.w_forceRef)

        #######################
        # FOOT TRACKING TASKS #
        #######################

        self.feetTask = 4*[None]  # List to store the foot tracking tasks
        mask = np.matrix([1.0, 1.0, 1.0, 0.0, 0.0, 0.0]).T

        # Task definition (creating the task object)
        for i_foot in range(4):
            self.feetTask[i_foot] = tsid.TaskSE3Equality(
                "foot_track_" + str(i_foot), self.robot, self.foot_frames[i_foot])
            self.feetTask[i_foot].setKp(kp_foot * mask)
            self.feetTask[i_foot].setKd(2.0 * np.sqrt(kp_foot) * mask)
            self.feetTask[i_foot].setMask(mask)
            self.feetTask[i_foot].useLocalFrame(False)

        # The reference will be set later when the task is enabled

        ##########
        # SOLVER #
        ##########

        # Use EiquadprogFast solver
        self.solver = tsid.SolverHQuadProgFast("qp solver")

        # Resize the solver to fit the number of variables, equality and inequality constraints
        self.solver.resize(self.invdyn.nVar, self.invdyn.nEq, self.invdyn.nIn)
	def __init__(self, q0, omega, t):
		
		self.qdes = q0.copy()
		self.vdes = np.zeros((8,1))
		self.ades = np.zeros((8,1))
		self.error = False
		
		kp_posture = 1.0
		w_posture = 10.0
		
		########################################################################
		#             Definition of the Model and TSID problem                 #
		########################################################################
		
		## Set the paths where the urdf and srdf file of the robot are registered

		modelPath = "/opt/openrobots/lib/python3.5/site-packages/../../../share/example-robot-data/robots"
		urdf = modelPath + "/solo_description/robots/solo.urdf"
		srdf = modelPath + "/solo_description/srdf/solo.srdf"
		vector = pin.StdVec_StdString()
		vector.extend(item for item in modelPath)

		## Create the robot wrapper from the urdf model (without the free flyer)

		self.robot = tsid.RobotWrapper(urdf, vector, False)
		
		self.model = self.robot.model()
		
		## Creation of the Invverse Dynamics HQP problem using the robot
		## accelerations (base + joints) and the contact forces

		self.invdyn = tsid.InverseDynamicsFormulationAccForce("tsid", self.robot, False)
		# Compute the problem data with a solver based on EiQuadProg
		self.invdyn.computeProblemData(t, self.qdes, self.vdes)
		# Get the initial data
		self.data = self.invdyn.data()
		
		## Task definition
		
		# POSTURE Task
		self.postureTask = tsid.TaskJointPosture("task-posture", self.robot)
		self.postureTask.setKp(kp_posture * matlib.ones(8).T) # Proportional gain 
		self.postureTask.setKd(2.0 * np.sqrt(kp_posture) * matlib.ones(8).T) # Derivative gain 
		# Add the task to the HQP with weight = w_posture, priority level = 0 (as real constraint) and a transition duration = 0.0
		self.invdyn.addMotionTask(self.postureTask, w_posture, 0, 0.0)

		
		## TSID Trajectory 
		
		pin.loadReferenceConfigurations(self.model, srdf, False)
		
		q_ref = self.model.referenceConfigurations['straight_standing'] 
		trajPosture = tsid.TrajectoryEuclidianConstant("traj_joint", q_ref)
		
		# Set the trajectory as reference for the posture task
		samplePosture = trajPosture.computeNext()
		self.postureTask.setReference(samplePosture)
		
		## Initialization of the solver

		# Use EiquadprogFast solver
		self.solver = tsid.SolverHQuadProgFast("qp solver")
		# Resize the solver to fit the number of variables, equality and inequality constraints
		self.solver.resize(self.invdyn.nVar, self.invdyn.nEq, self.invdyn.nIn)
Exemple #8
0
vector = se3.StdVec_StdString()
vector.extend(item for item in conf.path)
robot = tsid.RobotWrapper(conf.urdf, vector, False)
model = robot.model()

formulation = tsid.InverseDynamicsFormulationAccForce("tsid", robot, False)
q0 = conf.q0
v0 = np.array(np.zeros(robot.nv)).T
formulation.computeProblemData(0.0, q0, v0)

postureTask = tsid.TaskJointPosture("task-posture", robot)
postureTask.setKp(conf.kp_posture * np.ones(robot.nv).T)
postureTask.setKd(2.0 * np.sqrt(conf.kp_posture) * np.ones(robot.nv).T)
formulation.addMotionTask(postureTask, conf.w_posture, 1, 0.0)

trajPosture = tsid.TrajectoryEuclidianConstant("traj_joint", q0)
postureTask.setReference(trajPosture.computeNext())

v_max = conf.v_max_scaling * model.velocityLimit
v_min = -v_max
jointBoundsTask = tsid.TaskJointBounds("task-joint-bounds", robot, conf.dt)
jointBoundsTask.setVelocityBounds(v_min, v_max)
formulation.addMotionTask(jointBoundsTask, conf.w_joint_bounds, 0, 0.0)

solver = tsid.SolverHQuadProgFast("qp solver")
solver.resize(formulation.nVar, formulation.nEq, formulation.nIn)

if (USE_VIEWER):
    robot_display = se3.RobotWrapper.BuildFromURDF(conf.urdf, [
        conf.path,
    ])
def generateWholeBodyMotion(cs, fullBody=None, viewer=None):
    if not viewer:
        print "No viewer linked, cannot display end_effector trajectories."
    print "Start TSID ... "

    rp = RosPack()
    urdf = rp.get_path(
        cfg.Robot.packageName
    ) + '/urdf/' + cfg.Robot.urdfName + cfg.Robot.urdfSuffix + '.urdf'
    if cfg.WB_VERBOSE:
        print "load robot : ", urdf
    #srdf = "package://" + package + '/srdf/' +  cfg.Robot.urdfName+cfg.Robot.srdfSuffix + '.srdf'
    robot = tsid.RobotWrapper(urdf, pin.StdVec_StdString(),
                              pin.JointModelFreeFlyer(), False)
    if cfg.WB_VERBOSE:
        print "robot loaded in tsid."
    # FIXME : tsid robotWrapper don't have all the required methods, only pinocchio have them
    pinRobot = pin.RobotWrapper.BuildFromURDF(urdf, pin.StdVec_StdString(),
                                              pin.JointModelFreeFlyer(), False)

    q = cs.contact_phases[0].reference_configurations[0][:robot.nq].copy()
    v = np.matrix(np.zeros(robot.nv)).transpose()
    t = 0.0  # time
    # init states list with initial state (assume joint velocity is null for t=0)
    invdyn = tsid.InverseDynamicsFormulationAccForce("tsid", robot, False)
    invdyn.computeProblemData(t, q, v)
    data = invdyn.data()

    if cfg.EFF_CHECK_COLLISION:  # initialise object needed to check the motion
        from mlp.utils import check_path
        validator = check_path.PathChecker(fullBody, cs, len(q),
                                           cfg.WB_VERBOSE)

    if cfg.WB_VERBOSE:
        print "initialize tasks : "
    comTask = tsid.TaskComEquality("task-com", robot)
    comTask.setKp(cfg.kp_com * np.matrix(np.ones(3)).transpose())
    comTask.setKd(2.0 * np.sqrt(cfg.kp_com) *
                  np.matrix(np.ones(3)).transpose())
    invdyn.addMotionTask(comTask, cfg.w_com, cfg.level_com, 0.0)

    com_ref = robot.com(invdyn.data())
    trajCom = tsid.TrajectoryEuclidianConstant("traj_com", com_ref)

    amTask = tsid.TaskAMEquality("task-am", robot)
    amTask.setKp(cfg.kp_am * np.matrix([1., 1., 0.]).T)
    amTask.setKd(2.0 * np.sqrt(cfg.kp_am * np.matrix([1., 1., 0.]).T))
    invdyn.addTask(amTask, cfg.w_am, cfg.level_am)
    trajAM = tsid.TrajectoryEuclidianConstant("traj_am",
                                              np.matrix(np.zeros(3)).T)

    postureTask = tsid.TaskJointPosture("task-joint-posture", robot)
    postureTask.setKp(cfg.kp_posture * cfg.gain_vector)
    postureTask.setKd(2.0 * np.sqrt(cfg.kp_posture * cfg.gain_vector))
    postureTask.mask(cfg.masks_posture)
    invdyn.addMotionTask(postureTask, cfg.w_posture, cfg.level_posture, 0.0)
    q_ref = cfg.IK_REFERENCE_CONFIG
    trajPosture = tsid.TrajectoryEuclidianConstant("traj_joint", q_ref[7:])

    orientationRootTask = tsid.TaskSE3Equality("task-orientation-root", robot,
                                               'root_joint')
    mask = np.matrix(np.ones(6)).transpose()
    mask[0:3] = 0
    mask[5] = cfg.YAW_ROT_GAIN
    orientationRootTask.setKp(cfg.kp_rootOrientation * mask)
    orientationRootTask.setKd(2.0 * np.sqrt(cfg.kp_rootOrientation * mask))
    invdyn.addMotionTask(orientationRootTask, cfg.w_rootOrientation,
                         cfg.level_rootOrientation, 0.0)
    root_ref = robot.position(data, robot.model().getJointId('root_joint'))
    trajRoot = tsid.TrajectorySE3Constant("traj-root", root_ref)

    usedEffectors = []
    for eeName in cfg.Robot.dict_limb_joint.values():
        if isContactEverActive(cs, eeName):
            usedEffectors.append(eeName)
    # init effector task objects :
    dic_effectors_tasks = createEffectorTasksDic(cs, robot)
    effectorTraj = tsid.TrajectorySE3Constant(
        "traj-effector", SE3.Identity()
    )  # trajectory doesn't matter as it's only used to get the correct struct and size

    # init empty dic to store effectors trajectories :
    dic_effectors_trajs = {}
    for eeName in usedEffectors:
        dic_effectors_trajs.update({eeName: None})

    # add initial contacts :
    dic_contacts = {}
    for eeName in usedEffectors:
        if isContactActive(cs.contact_phases[0], eeName):
            contact = createContactForEffector(invdyn, robot,
                                               cs.contact_phases[0], eeName)
            dic_contacts.update({eeName: contact})

    if cfg.PLOT:  # init a dict storing all the reference trajectories used (for plotting)
        stored_effectors_ref = {}
        for eeName in dic_effectors_tasks:
            stored_effectors_ref.update({eeName: []})

    solver = tsid.SolverHQuadProg("qp solver")
    solver.resize(invdyn.nVar, invdyn.nEq, invdyn.nIn)

    # define nested function used in control loop
    def storeData(k_t, res, q, v, dv, invdyn, sol):
        # store current state
        res.q_t[:, k_t] = q
        res.dq_t[:, k_t] = v
        res.ddq_t[:, k_t] = dv
        res.tau_t[:, k_t] = invdyn.getActuatorForces(
            sol)  # actuator forces, with external forces (contact forces)
        #store contact info (force and status)
        if cfg.IK_store_contact_forces:
            for eeName, contact in dic_contacts.iteritems():
                if invdyn.checkContact(contact.name, sol):
                    res.contact_forces[eeName][:,
                                               k_t] = invdyn.getContactForce(
                                                   contact.name, sol)
                    res.contact_normal_force[
                        eeName][:, k_t] = contact.getNormalForce(
                            res.contact_forces[eeName][:, k_t])
                    res.contact_activity[eeName][:, k_t] = 1
        # store centroidal info (real one and reference) :
        if cfg.IK_store_centroidal:
            pcom, vcom, acom = pinRobot.com(q, v, dv)
            res.c_t[:, k_t] = pcom
            res.dc_t[:, k_t] = vcom
            res.ddc_t[:, k_t] = acom
            res.L_t[:, k_t] = pinRobot.centroidalMomentum(q, v).angular
            #res.dL_t[:,k_t] = pinRobot.centroidalMomentumVariation(q,v,dv) # FIXME : in robot wrapper, use * instead of .dot() for np matrices
            pin.dccrba(pinRobot.model, pinRobot.data, q, v)
            res.dL_t[:, k_t] = Force(
                pinRobot.data.Ag.dot(dv) + pinRobot.data.dAg.dot(v)).angular
            # same for reference data :
            res.c_reference[:, k_t] = com_desired
            res.dc_reference[:, k_t] = vcom_desired
            res.ddc_reference[:, k_t] = acom_desired
            res.L_reference[:, k_t] = L_desired
            res.dL_reference[:, k_t] = dL_desired
            if cfg.IK_store_zmp:
                tau = pin.rnea(
                    pinRobot.model, pinRobot.data, q, v, dv
                )  # tau without external forces, only used for the 6 first
                #res.tau_t[:6,k_t] = tau[:6]
                phi0 = pinRobot.data.oMi[1].act(Force(tau[:6]))
                res.wrench_t[:, k_t] = phi0.vector
                res.zmp_t[:, k_t] = shiftZMPtoFloorAltitude(
                    cs, res.t_t[k_t], phi0, cfg.EXPORT_OPENHRP)
                # same but from the 'reference' values :
                Mcom = SE3.Identity()
                Mcom.translation = com_desired
                Fcom = Force.Zero()
                Fcom.linear = cfg.MASS * (acom_desired - cfg.GRAVITY)
                Fcom.angular = dL_desired
                F0 = Mcom.act(Fcom)
                res.wrench_reference[:, k_t] = F0.vector
                res.zmp_reference[:, k_t] = shiftZMPtoFloorAltitude(
                    cs, res.t_t[k_t], F0, cfg.EXPORT_OPENHRP)
        if cfg.IK_store_effector:
            for eeName in usedEffectors:  # real position (not reference)
                res.effector_trajectories[eeName][:, k_t] = SE3toVec(
                    getCurrentEffectorPosition(robot, invdyn.data(), eeName))
                res.d_effector_trajectories[eeName][:, k_t] = MotiontoVec(
                    getCurrentEffectorVelocity(robot, invdyn.data(), eeName))
                res.dd_effector_trajectories[eeName][:, k_t] = MotiontoVec(
                    getCurrentEffectorAcceleration(robot, invdyn.data(),
                                                   eeName))

        return res

    def printIntermediate(v, dv, invdyn, sol):
        print "Time %.3f" % (t)
        for eeName, contact in dic_contacts.iteritems():
            if invdyn.checkContact(contact.name, sol):
                f = invdyn.getContactForce(contact.name, sol)
                print "\tnormal force %s: %.1f" % (contact.name.ljust(
                    20, '.'), contact.getNormalForce(f))

        print "\ttracking err %s: %.3f" % (comTask.name.ljust(
            20, '.'), norm(comTask.position_error, 2))
        for eeName, traj in dic_effectors_trajs.iteritems():
            if traj:
                task = dic_effectors_tasks[eeName]
                error = task.position_error
                if cfg.Robot.cType == "_3_DOF":
                    error = error[0:3]
                print "\ttracking err %s: %.3f" % (task.name.ljust(
                    20, '.'), norm(error, 2))
        print "\t||v||: %.3f\t ||dv||: %.3f" % (norm(v, 2), norm(dv))

    def checkDiverge(res, v, dv):
        if norm(dv) > 1e6 or norm(v) > 1e6:
            print "!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!"
            print "/!\ ABORT : controler unstable at t = " + str(t) + "  /!\ "
            print "!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!"
            raise ValueError("ABORT : controler unstable at t = " + str(t))
        if math.isnan(norm(dv)) or math.isnan(norm(v)):
            print "!!!!!!    !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!"
            print "/!\ ABORT : nan   at t = " + str(t) + "  /!\ "
            print "!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!"
            raise ValueError("ABORT : controler unstable at t = " + str(t))

    def stopHere():
        if cfg.WB_ABORT_WHEN_INVALID:
            return res.resize(phase_interval[0]), pinRobot
        elif cfg.WB_RETURN_INVALID:
            return res.resize(k_t), pinRobot

    # time check
    dt = cfg.IK_dt
    if cfg.WB_VERBOSE:
        print "dt : ", dt
    res = Result(robot.nq, robot.nv, cfg.IK_dt, eeNames=usedEffectors, cs=cs)
    N = res.N
    last_display = 0
    if cfg.WB_VERBOSE:
        print "tsid initialized, start control loop"
        #raw_input("Enter to start the motion (motion displayed as it's computed, may be slower than real-time)")
    time_start = time.time()

    # For each phases, create the necessary task and references trajectories :
    for pid in range(cs.size()):
        if cfg.WB_VERBOSE:
            print "## for phase : ", pid
            print "t = ", t
        phase = cs.contact_phases[pid]
        if pid < cs.size() - 1:
            phase_next = cs.contact_phases[pid + 1]
        else:
            phase_next = None
        if pid > 0:
            phase_prev = cs.contact_phases[pid - 1]
        else:
            phase_prev = None
        t_phase_begin = res.phases_intervals[pid][0] * dt
        t_phase_end = res.phases_intervals[pid][-1] * dt
        time_interval = [t_phase_begin, t_phase_end]
        if cfg.WB_VERBOSE:
            print "time_interval ", time_interval
        # generate com ref traj from phase :
        com_init = np.matrix(np.zeros((9, 1)))
        com_init[0:3, 0] = robot.com(invdyn.data())
        com_traj = computeCOMRefFromPhase(phase, time_interval)

        am_traj = computeAMRefFromPhase(phase, time_interval)
        # add root's orientation ref from reference config :
        if cfg.USE_PLANNING_ROOT_ORIENTATION:
            if phase_next:
                root_traj = trajectories.TrajectorySE3LinearInterp(
                    SE3FromConfig(phase.reference_configurations[0]),
                    SE3FromConfig(phase_next.reference_configurations[0]),
                    time_interval)
            else:
                root_traj = trajectories.TrajectorySE3LinearInterp(
                    SE3FromConfig(phase.reference_configurations[0]),
                    SE3FromConfig(phase.reference_configurations[0]),
                    time_interval)
        else:
            # orientation such that the torso orientation is the mean between both feet yaw rotations:
            placement_init, placement_end = rootOrientationFromFeetPlacement(
                phase, phase_next)
            root_traj = trajectories.TrajectorySE3LinearInterp(
                placement_init, placement_end, time_interval)
        # add newly created contacts :
        for eeName in usedEffectors:
            if phase_prev and not isContactActive(
                    phase_prev, eeName) and isContactActive(phase, eeName):
                invdyn.removeTask(dic_effectors_tasks[eeName].name,
                                  0.0)  # remove pin task for this contact
                dic_effectors_trajs.update(
                    {eeName:
                     None})  # delete reference trajectory for this task
                if cfg.WB_VERBOSE:
                    print "remove se3 task : " + dic_effectors_tasks[
                        eeName].name
                contact = createContactForEffector(invdyn, robot, phase,
                                                   eeName)
                dic_contacts.update({eeName: contact})

        # add se3 tasks for end effector not in contact that will be in contact next phase:
        for eeName, task in dic_effectors_tasks.iteritems():
            if phase_next and not isContactActive(
                    phase, eeName) and isContactActive(phase_next, eeName):
                if cfg.WB_VERBOSE:
                    print "add se3 task for " + eeName
                invdyn.addMotionTask(task, cfg.w_eff, cfg.level_eff, 0.0)
                #create reference trajectory for this task :
                placement_init = getCurrentEffectorPosition(
                    robot, invdyn.data(), eeName
                )  #FIXME : adjust orientation in case of 3D contact ...
                if cfg.Robot.cType == "_3_DOF":
                    placement_init.rotation = JointPlacementForEffector(
                        phase, eeName).rotation
                placement_end = JointPlacementForEffector(phase_next, eeName)
                ref_traj = generateEndEffectorTraj(time_interval,
                                                   placement_init,
                                                   placement_end, 0)
                if cfg.WB_VERBOSE:
                    print "t interval : ", time_interval
                    print "placement Init : ", placement_init
                    print "placement End  : ", placement_end
                    # display all the effector trajectories for this phase
                if viewer and cfg.DISPLAY_ALL_FEET_TRAJ:
                    display_tools.displaySE3Traj(
                        ref_traj, viewer.client.gui, viewer.sceneName,
                        eeName + "_traj_" + str(pid),
                        cfg.Robot.dict_limb_color_traj[eeName], time_interval,
                        cfg.Robot.dict_offset[eeName])
                    viewer.client.gui.setVisibility(
                        eeName + "_traj_" + str(pid), 'ALWAYS_ON_TOP')
                dic_effectors_trajs.update({eeName: ref_traj})

        # start removing the contact that will be broken in the next phase :
        # (This tell the solver that it should start minimzing the contact force on this contact, and ideally get to 0 at the given time)
        for eeName, contact in dic_contacts.iteritems():
            if phase_next and isContactActive(
                    phase, eeName) and not isContactActive(phase_next, eeName):
                transition_time = t_phase_end - t - dt / 2.
                if cfg.WB_VERBOSE:
                    print "\nTime %.3f Start breaking contact %s. transition time : %.3f\n" % (
                        t, contact.name, transition_time)
                invdyn.removeRigidContact(contact.name, transition_time)

        if cfg.WB_STOP_AT_EACH_PHASE:
            raw_input('start simulation')
        # save values at the beginning of the current phase
        q_begin = q.copy()
        v_begin = v.copy()
        phaseValid = False
        swingPhase = False  # will be true if an effector move during this phase
        iter_for_phase = -1
        # iterate until a valid motion for this phase is found (ie. collision free and which respect joint-limits)
        while not phaseValid:
            if iter_for_phase >= 0:
                # reset values to their value at the beginning of the current phase
                q = q_begin.copy()
                v = v_begin.copy()
            iter_for_phase += 1
            if cfg.WB_VERBOSE:
                print "Start simulation for phase " + str(
                    pid) + ", try number :  " + str(iter_for_phase)
            # loop to generate states (q,v,a) for the current contact phase :
            if pid == cs.size() - 1:  # last state
                phase_interval = res.phases_intervals[pid]
            else:
                phase_interval = res.phases_intervals[pid][:-1]
            for k_t in phase_interval:
                t = res.t_t[k_t]
                # set traj reference for current time :
                # com
                sampleCom = trajCom.computeNext()
                com_desired = com_traj(t)[0]
                vcom_desired = com_traj(t)[1]
                acom_desired = com_traj(t)[2]
                sampleCom.pos(com_desired)
                sampleCom.vel(vcom_desired)
                sampleCom.acc(acom_desired)
                comTask.setReference(sampleCom)

                # am
                sampleAM = trajAM.computeNext()
                L_desired = am_traj(t)[0]
                dL_desired = am_traj(t)[1]
                sampleAM.pos(L_desired)
                sampleAM.vel(dL_desired)
                amTask.setReference(sampleAM)

                # posture
                samplePosture = trajPosture.computeNext()
                #print "postural task ref : ",samplePosture.pos()
                postureTask.setReference(samplePosture)

                # root orientation :
                sampleRoot = trajRoot.computeNext()
                sampleRoot.pos(SE3toVec(root_traj(t)[0]))
                sampleRoot.vel(MotiontoVec(root_traj(t)[1]))

                orientationRootTask.setReference(sampleRoot)
                quat_waist = Quaternion(root_traj(t)[0].rotation)
                res.waist_orientation_reference[:, k_t] = np.matrix(
                    [quat_waist.x, quat_waist.y, quat_waist.z, quat_waist.w]).T
                if cfg.WB_VERBOSE == 2:
                    print "### references given : ###"
                    print "com  pos : ", sampleCom.pos()
                    print "com  vel : ", sampleCom.vel()
                    print "com  acc : ", sampleCom.acc()
                    print "AM   pos : ", sampleAM.pos()
                    print "AM   vel : ", sampleAM.vel()
                    print "root pos : ", sampleRoot.pos()
                    print "root vel : ", sampleRoot.vel()

                # end effector (if they exists)
                for eeName, traj in dic_effectors_trajs.iteritems():
                    if traj:
                        swingPhase = True  # there is an effector motion in this phase
                        sampleEff = effectorTraj.computeNext()
                        traj_t = traj(t)
                        sampleEff.pos(SE3toVec(traj_t[0]))
                        sampleEff.vel(MotiontoVec(traj_t[1]))
                        dic_effectors_tasks[eeName].setReference(sampleEff)
                        if cfg.WB_VERBOSE == 2:
                            print "effector " + str(eeName) + " pos : " + str(
                                sampleEff.pos())
                            print "effector " + str(eeName) + " vel : " + str(
                                sampleEff.vel())
                        if cfg.IK_store_effector:
                            res.effector_references[eeName][:, k_t] = SE3toVec(
                                traj_t[0])
                            res.d_effector_references[
                                eeName][:, k_t] = MotiontoVec(traj_t[1])
                            res.dd_effector_references[
                                eeName][:, k_t] = MotiontoVec(traj_t[2])
                    elif cfg.IK_store_effector:
                        if k_t == 0:
                            res.effector_references[eeName][:, k_t] = SE3toVec(
                                getCurrentEffectorPosition(
                                    robot, invdyn.data(), eeName))
                        else:
                            res.effector_references[
                                eeName][:, k_t] = res.effector_references[
                                    eeName][:, k_t - 1]
                        res.d_effector_references[eeName][:, k_t] = np.matrix(
                            np.zeros(6)).T
                        res.dd_effector_references[eeName][:, k_t] = np.matrix(
                            np.zeros(6)).T

                # solve HQP for the current time
                HQPData = invdyn.computeProblemData(t, q, v)
                if cfg.WB_VERBOSE and t < phase.time_trajectory[0] + dt:
                    print "final data for phase ", pid
                    HQPData.print_all()

                sol = solver.solve(HQPData)
                dv = invdyn.getAccelerations(sol)
                res = storeData(k_t, res, q, v, dv, invdyn, sol)
                # update state
                v_mean = v + 0.5 * dt * dv
                v += dt * dv
                q = pin.integrate(robot.model(), q, dt * v_mean)

                if cfg.WB_VERBOSE == 2:
                    print "v = ", v
                    print "dv = ", dv

                if cfg.WB_VERBOSE and int(t / dt) % cfg.IK_PRINT_N == 0:
                    printIntermediate(v, dv, invdyn, sol)
                try:
                    checkDiverge(res, v, dv)
                except ValueError:
                    return stopHere()

            # end while t \in phase_t (loop for the current contact phase)
            if swingPhase and cfg.EFF_CHECK_COLLISION:
                phaseValid, t_invalid = validator.check_motion(
                    res.q_t[:, phase_interval[0]:k_t])
                #if iter_for_phase < 3 :# FIXME : debug only, force limb-rrt
                #    phaseValid = False
                if not phaseValid:
                    print "Phase " + str(pid) + " not valid at t = " + str(
                        t_invalid)
                    if t_invalid <= cfg.EFF_T_PREDEF or t_invalid >= (
                        (t_phase_end - t_phase_begin) - cfg.EFF_T_PREDEF):
                        print "Motion is invalid during predefined phases, cannot change this."
                        return stopHere()
                    if effectorCanRetry():
                        print "Try new end effector trajectory."
                        try:
                            for eeName, oldTraj in dic_effectors_trajs.iteritems(
                            ):
                                if oldTraj:  # update the traj in the map
                                    placement_init = JointPlacementForEffector(
                                        phase_prev, eeName)
                                    placement_end = JointPlacementForEffector(
                                        phase_next, eeName)
                                    ref_traj = generateEndEffectorTraj(
                                        time_interval, placement_init,
                                        placement_end, iter_for_phase + 1,
                                        res.q_t[:, phase_interval[0]:k_t],
                                        phase_prev, phase, phase_next,
                                        fullBody, eeName, viewer)
                                    dic_effectors_trajs.update(
                                        {eeName: ref_traj})
                                    if viewer and cfg.DISPLAY_ALL_FEET_TRAJ:
                                        display_tools.displaySE3Traj(
                                            ref_traj, viewer.client.gui,
                                            viewer.sceneName,
                                            eeName + "_traj_" + str(pid), cfg.
                                            Robot.dict_limb_color_traj[eeName],
                                            time_interval,
                                            cfg.Robot.dict_offset[eeName])
                                        viewer.client.gui.setVisibility(
                                            eeName + "_traj_" + str(pid),
                                            'ALWAYS_ON_TOP')
                        except ValueError, e:
                            print "ERROR in generateEndEffectorTraj :"
                            print e.message
                            return stopHere()
                    else:
                        print "End effector method choosen do not allow retries, abort here."
                        return stopHere()
            else:  # no effector motions, phase always valid (or bypass the check)
                phaseValid = True
                if cfg.WB_VERBOSE:
                    print "Phase " + str(pid) + " valid."
            if phaseValid:
                # display all the effector trajectories for this phase
                if viewer and cfg.DISPLAY_FEET_TRAJ and not cfg.DISPLAY_ALL_FEET_TRAJ:
                    for eeName, ref_traj in dic_effectors_trajs.iteritems():
                        if ref_traj:
                            display_tools.displaySE3Traj(
                                ref_traj, viewer.client.gui, viewer.sceneName,
                                eeName + "_traj_" + str(pid),
                                cfg.Robot.dict_limb_color_traj[eeName],
                                time_interval, cfg.Robot.dict_offset[eeName])
                            viewer.client.gui.setVisibility(
                                eeName + "_traj_" + str(pid), 'ALWAYS_ON_TOP')
                            if cfg.PLOT:  # add current ref_traj to the list for plotting
                                stored_effectors_ref[eeName] += [ref_traj]
Exemple #10
0
print "q:", q.transpose()

se3.centerOfMass(model, data, q)

taskCOM = tsid.TaskComEquality("task-com", robot)

Kp = 100 * np.matrix(np.ones(3)).transpose()
Kd = 20.0 * np.matrix(np.ones(3)).transpose()
taskCOM.setKp(Kp)
taskCOM.setKd(Kd)

assert np.linalg.norm(Kp - taskCOM.Kp, 2) < tol
assert np.linalg.norm(Kd - taskCOM.Kd, 2) < tol

com_ref = data.com[0] + np.matrix(np.ones(3) * 0.02).transpose()
traj = tsid.TrajectoryEuclidianConstant("traj_se3", com_ref)
sample = tsid.TrajectorySample(0)

t = 0.0
dt = 0.001
max_it = 1000
Jpinv = np.matrix(np.zeros((robot.nv, 3)))
error_past = 1e100
v = np.matrix(np.zeros(robot.nv)).transpose()

for i in range(0, max_it):
    robot.computeAllTerms(data, q, v)
    sample = traj.computeNext()
    taskCOM.setReference(sample)
    const = taskCOM.compute(t, q, v, data)
    def __init__(self, q0, omega, t):

        self.omega = omega
        self.qdes = q0.copy()
        self.vdes = np.zeros((18, 1))
        self.ades = np.zeros((18, 1))
        self.error = False

        kp_posture = 10.0  # proportionnal gain of the posture task
        w_posture = 1.0  # weight of the posture task

        kp_com = 100.0  # proportionnal gain of the CoM task
        w_com = 100.0  # weight of the CoM task

        kp_lock = 1.0  # proportionnal gain of the lock task
        w_lock = 100.0  # weight of the lock task

        # For the contacts
        mu = 0.3  # friction coefficient
        fMin = 1.0  # minimum normal force
        fMax = 100.0  # maximum normal force

        w_forceRef = 1e-3  # weight of the forces regularization
        kp_contact = 1.0  # proportionnal gain for the contacts

        foot_frames = ['HL_FOOT', 'HR_FOOT', 'FL_FOOT',
                       'FR_FOOT']  # tab with all the foot frames names
        contactNormal = np.matrix(
            [0., 0., 1.]).T  # direction of the normal to the contact surface

        ########################################################################
        #             Definition of the Model and TSID problem                 #
        ########################################################################

        ## Set the paths where the urdf and srdf file of the robot are registered

        modelPath = "/opt/openrobots/lib/python3.5/site-packages/../../../share/example-robot-data/robots"
        urdf = modelPath + "/solo_description/robots/solo12.urdf"
        srdf = modelPath + "/solo_description/srdf/solo.srdf"
        vector = pin.StdVec_StdString()
        vector.extend(item for item in modelPath)

        ## Create the robot wrapper from the urdf model (without the free flyer)

        self.robot = tsid.RobotWrapper(urdf, vector, pin.JointModelFreeFlyer(),
                                       False)

        self.model = self.robot.model()

        ## Creation of the Invverse Dynamics HQP problem using the robot
        ## accelerations (base + joints) and the contact forces

        self.invdyn = tsid.InverseDynamicsFormulationAccForce(
            "tsid", self.robot, False)
        # Compute the problem data with a solver based on EiQuadProg
        self.invdyn.computeProblemData(t, self.qdes, self.vdes)
        # Get the initial data
        self.data = self.invdyn.data()

        ## Task definition

        # POSTURE Task
        self.postureTask = tsid.TaskJointPosture("task-posture", self.robot)
        self.postureTask.setKp(
            kp_posture * matlib.ones(self.robot.nv - 6).T)  # Proportional gain
        self.postureTask.setKd(
            2.0 * np.sqrt(kp_posture) *
            matlib.ones(self.robot.nv - 6).T)  # Derivative gain
        # Add the task to the HQP with weight = w_posture, priority level = 0 (as real constraint) and a transition duration = 0.0
        self.invdyn.addMotionTask(self.postureTask, w_posture, 1, 0.0)

        # COM Task
        self.comTask = tsid.TaskComEquality("task-com", self.robot)
        self.comTask.setKp(
            kp_com * matlib.ones(3).T)  # Proportional gain of the CoM task
        self.comTask.setKd(2.0 * np.sqrt(kp_com) *
                           matlib.ones(3).T)  # Derivative gain
        self.invdyn.addMotionTask(
            self.comTask, w_com, 1, 0.0
        )  # Add the task to the HQP with weight = w_com, priority level = 1 (in the cost function) and a transition duration = 0.0

        # LOCK Task
        self.lockTask = tsid.TaskJointPosture("task-lock-shoulder", self.robot)
        self.lockTask.setKp(kp_lock * matlib.ones(self.robot.nv - 6).T)
        self.lockTask.setKd(2.0 * np.sqrt(kp_lock) *
                            matlib.ones(self.robot.nv - 6).T)
        mask = np.matrix(np.zeros(
            self.robot.nv -
            6))  # Add a mask to take into account only the shoulders joints
        for i in [0, 3, 6, 9]:
            mask[0, i] = 1
        self.lockTask.mask(mask.T)
        self.invdyn.addMotionTask(
            self.lockTask, w_lock, 0,
            0.0)  # Add the task as real constraint (priority level = 0)

        ## CONTACTS

        self.contacts = 4 * [None]

        for i, name in enumerate(foot_frames):
            self.contacts[i] = tsid.ContactPoint(name, self.robot, name,
                                                 contactNormal, mu, fMin, fMax)
            self.contacts[i].setKp(kp_contact * matlib.ones(3).T)
            self.contacts[i].setKd(2.0 * np.sqrt(kp_contact) *
                                   matlib.ones(3).T)
            H_ref = self.robot.framePosition(self.data,
                                             self.model.getFrameId(name))
            self.contacts[i].setReference(H_ref)
            self.contacts[i].useLocalFrame(False)
            self.invdyn.addRigidContact(self.contacts[i], w_forceRef, 1.0, 1)

        ## TSID Trajectory

        # POSTURE Task
        pin.loadReferenceConfigurations(self.model, srdf, False)

        self.q_ref = self.model.referenceConfigurations['straight_standing']
        self.trajPosture = tsid.TrajectoryEuclidianConstant(
            "traj_joint", self.q_ref[7:])
        # Set the trajectory as reference for the posture task
        self.samplePosture = self.trajPosture.computeNext()
        self.postureTask.setReference(self.samplePosture)

        # COM Task
        self.com_ref = self.data.com[0].copy()  # Initial value of the CoM
        self.trajCom = tsid.TrajectoryEuclidianConstant(
            "traj_com", self.com_ref)

        self.sampleCom = self.trajCom.computeNext()
        self.comTask.setReference(self.sampleCom)

        # LOCK Task
        # The mask is enough to set the shoulder acceleration to 0 because 0 is the initial configuration for the shoulders
        trajLock = tsid.TrajectoryEuclidianConstant("traj_lock_shoulder",
                                                    self.q_ref[7:])

        sampleLock = trajLock.computeNext()
        self.lockTask.setReference(sampleLock)

        ## Initialization of the solver

        # Use EiquadprogFast solver
        self.solver = tsid.SolverHQuadProgFast("qp solver")
        # Resize the solver to fit the number of variables, equality and inequality constraints
        self.solver.resize(self.invdyn.nVar, self.invdyn.nEq, self.invdyn.nIn)
Exemple #12
0
    def __init__(self,
                 urdf_path,
                 model_path,
                 eff_arr,
                 q0,
                 v0,
                 mu=0.6,
                 fz_max=75):
        """
        Input:
            urdf_path: path to urdf of robot
            model_path: path to model of robot (required by TSID)
            eff_arr : end effector name array
            q0: initial configuration
            v0: initial velocity
            tau_max: vector of maximum torques for each joint
            mu: coefficient of friction at end-effectors
        """
        self.kp_com = 0.0001
        self.kp_contact = 5.0
        self.kp_posture = 10.0
        self.kp_orientation = 0.002

        self.w_com = 1.0
        self.w_posture = 1e0
        self.w_force = 1.0
        self.w_orientation = 1e-6

        self.contact_transition = 0.1

        self.eff_arr = eff_arr
        self.curr_cnt_array = np.zeros(len(self.eff_arr))

        #self.tau_max = tau_max
        #self.tau_min = -1*self.tau_max

        self.tsid_robot = tsid.RobotWrapper(urdf_path, [model_path],
                                            pin.JointModelFreeFlyer(), False)
        self.nq = self.tsid_robot.nq
        self.nv = self.tsid_robot.nv
        self.tsid_model = self.tsid_robot.model()
        self.invdyn = tsid.InverseDynamicsFormulationAccForce(
            "tsid", self.tsid_robot, False)
        self.qp_solver = tsid.SolverHQuadProgFast("qp_solver")

        #Set up initial solve
        self.invdyn.computeProblemData(0, q0.copy(), v0.copy())
        self.inv_dyn_data = self.invdyn.data()
        self.tsid_robot.computeAllTerms(self.inv_dyn_data, q0, v0)

        self.mu = mu

        # Add Contact Tasks (setup as equality constraints)
        contactNormal = np.array([0., 0., 1.])
        self.contact_arr = len(self.eff_arr) * [None]
        for i, name in enumerate(self.eff_arr):
            self.contact_arr[i] = tsid.ContactPoint(name, self.tsid_robot,
                                                    name, contactNormal,
                                                    self.mu, 0.01, fz_max)
            self.contact_arr[i].setKp(self.kp_contact * np.ones(3))
            self.contact_arr[i].setKd(.03 * np.ones(3))
            cnt_ref = self.tsid_robot.framePosition(
                self.inv_dyn_data,
                self.tsid_robot.model().getFrameId(name))
            self.contact_arr[i].setReference(cnt_ref)
            self.contact_arr[i].useLocalFrame(False)
            self.invdyn.addRigidContact(self.contact_arr[i], self.w_force)

        # Add posture tasks (in the cost function)
        self.postureTask = tsid.TaskJointPosture("posture_task",
                                                 self.tsid_robot)
        self.postureTask.setKp(self.kp_posture *
                               np.ones(self.tsid_robot.nv - 6))
        self.postureTask.setKd(.05 * np.ones(3))
        self.invdyn.addMotionTask(self.postureTask, self.w_posture, 1, 0.0)

        #Add CoM task (in the cost function)
        # self.comTask = tsid.TaskComEquality("task-com", self.tsid_robot)
        # self.comTask.setKp(self.kp_com * np.ones(6))
        # self.comTask.setKd(.01 * np.ones(6))
        # self.invdyn.addMotionTask(self.comTask, self.w_com, 1, 0.0)

        #Add orientation task (in the cost function)
        # self.oriTask = tsid.TaskSE3Equality("orientation", self.tsid_robot, "base_link")
        # self.oriTask.setKp(self.kp_orientation * np.ones(6))
        # self.oriTask.setKd(2.0 * np.sqrt(self.kp_orientation) * np.ones(6))
        # mask = np.ones(6)
        # mask[:3] = 0
        # self.oriTask.setMask(mask)
        # self.invdyn.addMotionTask(self.oriTask, self.w_orientation, 1, 0.0)

        #Setup trajectories (des_q, des_v, des_a)
        # self.com_ref = self.tsid_robot.com(self.inv_dyn_data)
        # self.trajCom = tsid.TrajectoryEuclidianConstant("trajectory_com", self.com_ref)
        # self.com_reference = self.trajCom.computeNext()
        #
        self.trajPosture = tsid.TrajectoryEuclidianConstant(
            "trajectory_posture", q0[7:])
        self.traj_reference = self.trajPosture.computeNext()
        self.postureTask.setReference(self.traj_reference)

        # self.ori_ref = self.tsid_robot.position(self.inv_dyn_data, self.tsid_robot.model().getJointId("base_link"))
        # self.oriTraj = tsid.TrajectorySE3Constant("trajectory_orientation", self.ori_ref)
        # self.ori_reference = self.oriTraj.computeNext()

        ### --- Actuation Bounds --- ###
        # Get actuator effort limits (i.e. torque limits) from URDF:
        # tau_max = self.actuator_scaling * self.tsid_model.effortLimit
        #actuationBounds = tsid.TaskActuationBounds("torque-actuation-bounds", self.tsid_robot)
        #actuationBounds.setBounds(self.tau_min, self.tau_max)
        #formulation.addActuationTask(actuationBounds, w_torque_bounds, 0, 0.0)

        self.qp_solver.resize(self.invdyn.nVar, self.invdyn.nEq,
                              self.invdyn.nIn)
Exemple #13
0
    def __init__(self, conf, dt, urdf, modelPath, srdf):
        self.conf = conf
        self.dt = dt
        vector = pin.StdVec_StdString()
        vector.extend(item for item in modelPath)
        self.robot = tsid.RobotWrapper(urdf, vector, pin.JointModelFreeFlyer(),
                                       False)
        robot = self.robot
        self.model = robot.model()
        pin.loadReferenceConfigurations(self.model, srdf, False)
        try:
            self.q0 = conf.q0
            q = np.copy(conf.q0)
        except:
            self.q0 = self.model.referenceConfigurations["half_sitting"]
            q = np.copy(self.q0)
        self.v0 = v = np.zeros(robot.nv)

        assert self.model.existFrame(conf.rf_frame_name)
        assert self.model.existFrame(conf.lf_frame_name)

        formulation = tsid.InverseDynamicsFormulationAccForce(
            "tsid", robot, False)
        formulation.computeProblemData(0.0, q, v)
        data = formulation.data()
        contact_Point = np.ones((3, 4)) * conf.lz
        contact_Point[0, :] = [-conf.lxn, -conf.lxn, conf.lxp, conf.lxp]
        contact_Point[1, :] = [-conf.lyn, conf.lyp, -conf.lyn, conf.lyp]

        contactRF = tsid.Contact6d("contact_rfoot", robot, conf.rf_frame_name,
                                   contact_Point, conf.contact_normal, conf.mu,
                                   conf.fMin, conf.fMax)
        contactRF.setKp(conf.kp_contact * np.ones(6))
        contactRF.setKd(2.0 * np.sqrt(conf.kp_contact) * np.ones(6))
        self.RF = robot.model().getFrameId(conf.rf_frame_name)
        H_rf_ref = robot.framePosition(formulation.data(), self.RF)
        #        print('H RF\n', H_rf_ref.translation)

        # modify initial robot configuration so that foot is on the ground (z=0)
        q[2] -= H_rf_ref.translation[2] + conf.lz
        formulation.computeProblemData(0.0, q, v)
        data = formulation.data()
        H_rf_ref = robot.framePosition(data, self.RF)
        #        print('H RF\n', H_rf_ref)
        contactRF.setReference(H_rf_ref)
        if (conf.w_contact >= 0.0):
            formulation.addRigidContact(contactRF, conf.w_forceRef,
                                        conf.w_contact, 1)
        else:
            formulation.addRigidContact(contactRF, conf.w_forceRef)

        contactLF = tsid.Contact6d("contact_lfoot", robot, conf.lf_frame_name,
                                   contact_Point, conf.contact_normal, conf.mu,
                                   conf.fMin, conf.fMax)
        contactLF.setKp(conf.kp_contact * np.ones(6))
        contactLF.setKd(2.0 * np.sqrt(conf.kp_contact) * np.ones(6))
        self.LF = robot.model().getFrameId(conf.lf_frame_name)
        H_lf_ref = robot.framePosition(formulation.data(), self.LF)
        #        print('H LF\n', H_lf_ref)
        contactLF.setReference(H_lf_ref)
        if (conf.w_contact >= 0.0):
            formulation.addRigidContact(contactLF, conf.w_forceRef,
                                        conf.w_contact, 1)
        else:
            formulation.addRigidContact(contactLF, conf.w_forceRef)

        comTask = tsid.TaskComEquality("task-com", robot)
        comTask.setKp(conf.kp_com * np.ones(3))
        comTask.setKd(2.0 * np.sqrt(conf.kp_com) * np.ones(3))
        if (conf.w_com > 0):
            formulation.addMotionTask(comTask, conf.w_com, 1, 0.0)

        postureTask = tsid.TaskJointPosture("task-posture", robot)
        try:
            postureTask.setKp(conf.kp_posture)
            postureTask.setKd(2.0 * np.sqrt(conf.kp_posture))
        except:
            postureTask.setKp(conf.kp_posture * np.ones(robot.nv - 6))
            postureTask.setKd(2.0 * np.sqrt(conf.kp_posture) *
                              np.ones(robot.nv - 6))
        if (conf.w_posture > 0):
            formulation.addMotionTask(postureTask, conf.w_posture, 1, 0.0)

        self.leftFootTask = tsid.TaskSE3Equality("task-left-foot", self.robot,
                                                 self.conf.lf_frame_name)
        self.leftFootTask.setKp(self.conf.kp_foot * np.ones(6))
        self.leftFootTask.setKd(2.0 * np.sqrt(self.conf.kp_foot) * np.ones(6))
        self.trajLF = tsid.TrajectorySE3Constant("traj-left-foot", H_lf_ref)
        if (conf.w_foot > 0):
            formulation.addMotionTask(self.leftFootTask, self.conf.w_foot, 1,
                                      0.0)
#
        self.rightFootTask = tsid.TaskSE3Equality("task-right-foot",
                                                  self.robot,
                                                  self.conf.rf_frame_name)
        self.rightFootTask.setKp(self.conf.kp_foot * np.ones(6))
        self.rightFootTask.setKd(2.0 * np.sqrt(self.conf.kp_foot) * np.ones(6))
        self.trajRF = tsid.TrajectorySE3Constant("traj-right-foot", H_rf_ref)
        if (conf.w_foot > 0):
            formulation.addMotionTask(self.rightFootTask, self.conf.w_foot, 1,
                                      0.0)

        self.waistTask = tsid.TaskSE3Equality("task-waist", self.robot,
                                              self.conf.waist_frame_name)
        self.waistTask.setKp(self.conf.kp_waist * np.ones(6))
        self.waistTask.setKd(2.0 * np.sqrt(self.conf.kp_waist) * np.ones(6))
        self.waistTask.setMask(np.array([0, 0, 0, 1, 1, 1.]))
        waistID = robot.model().getFrameId(conf.waist_frame_name)
        H_waist_ref = robot.framePosition(formulation.data(), waistID)
        self.trajWaist = tsid.TrajectorySE3Constant("traj-waist", H_waist_ref)
        if (conf.w_waist > 0):
            formulation.addMotionTask(self.waistTask, self.conf.w_waist, 1,
                                      0.0)

        self.tau_max = conf.tau_max_scaling * robot.model(
        ).effortLimit[-robot.na:]
        self.tau_min = -self.tau_max
        actuationBoundsTask = tsid.TaskActuationBounds("task-actuation-bounds",
                                                       robot)
        actuationBoundsTask.setBounds(self.tau_min, self.tau_max)
        if (conf.w_torque_bounds > 0.0):
            formulation.addActuationTask(actuationBoundsTask,
                                         conf.w_torque_bounds, 0, 0.0)

        jointBoundsTask = tsid.TaskJointBounds("task-joint-bounds", robot, dt)
        self.v_max = conf.v_max_scaling * robot.model(
        ).velocityLimit[-robot.na:]
        self.v_min = -self.v_max
        jointBoundsTask.setVelocityBounds(self.v_min, self.v_max)
        if (conf.w_joint_bounds > 0.0):
            formulation.addMotionTask(jointBoundsTask, conf.w_joint_bounds, 0,
                                      0.0)

        com_ref = robot.com(formulation.data())
        self.trajCom = tsid.TrajectoryEuclidianConstant("traj_com", com_ref)
        self.sample_com = self.trajCom.computeNext()

        q_ref = q[7:]
        self.trajPosture = tsid.TrajectoryEuclidianConstant(
            "traj_joint", q_ref)
        postureTask.setReference(self.trajPosture.computeNext())

        self.sampleLF = self.trajLF.computeNext()
        self.sample_LF_pos = self.sampleLF.pos()
        self.sample_LF_vel = self.sampleLF.vel()
        self.sample_LF_acc = self.sampleLF.acc()

        self.sampleRF = self.trajRF.computeNext()
        self.sample_RF_pos = self.sampleRF.pos()
        self.sample_RF_vel = self.sampleRF.vel()
        self.sample_RF_acc = self.sampleRF.acc()

        self.solver = tsid.SolverHQuadProgFast("qp solver")
        self.solver.resize(formulation.nVar, formulation.nEq, formulation.nIn)

        self.comTask = comTask
        self.postureTask = postureTask
        self.contactRF = contactRF
        self.contactLF = contactLF
        self.actuationBoundsTask = actuationBoundsTask
        self.jointBoundsTask = jointBoundsTask
        self.formulation = formulation
        self.q = q
        self.v = v
        self.q0 = np.copy(self.q)

        self.contact_LF_active = True
        self.contact_RF_active = True

        #        self.reset()

        data = np.load(conf.DATA_FILE_TSID)
        # assume dt>dt_ref
        r = int(dt / conf.dt_ref)

        self.N = int(data['com'].shape[1] / r)

        self.contact_phase = data['contact_phase'][::r]
        self.com_pos_ref = np.asarray(data['com'])[:, ::r]
        self.com_vel_ref = np.asarray(data['dcom'])[:, ::r]
        self.com_acc_ref = np.asarray(data['ddcom'])[:, ::r]
        self.x_RF_ref = np.asarray(data['x_RF'])[:, ::r]
        self.dx_RF_ref = np.asarray(data['dx_RF'])[:, ::r]
        self.ddx_RF_ref = np.asarray(data['ddx_RF'])[:, ::r]
        self.x_LF_ref = np.asarray(data['x_LF'])[:, ::r]
        self.dx_LF_ref = np.asarray(data['dx_LF'])[:, ::r]
        self.ddx_LF_ref = np.asarray(data['ddx_LF'])[:, ::r]
        self.cop_ref = np.asarray(data['cop'])[:, ::r]

        x_rf = self.get_placement_RF().translation
        offset = x_rf - self.x_RF_ref[:, 0]
        #        print("offset", offset)
        for i in range(self.N):
            self.com_pos_ref[:, i] += offset
            self.x_RF_ref[:, i] += offset
            self.x_LF_ref[:, i] += offset
Exemple #14
0
    def __init__(self, conf, viewer=pin.visualize.MeshcatVisualizer):
        self.conf = conf
        self.t_limit = False
        self.j_limit = True

        self.robot = tsid.RobotWrapper(conf.urdf, [conf.path],
                                       pin.JointModelFreeFlyer(), False)
        robot = self.robot
        self.model = robot.model()
        pin.loadReferenceConfigurations(self.model, conf.srdf, False)
        self.q0 = q = self.model.referenceConfigurations["half_sitting"]

        self.q0[0] = q[0] + 0.003 + 0.6
        self.q0[1] = q[1] - 0.001
        # self.q0[2] += 0.84

        q = self.q0
        v = np.zeros(robot.nv)

        assert self.model.existFrame(conf.rf_frame_name)
        assert self.model.existFrame(conf.lf_frame_name)

        formulation = tsid.InverseDynamicsFormulationAccForce(
            "tsid", robot, False)
        formulation.computeProblemData(0.0, q, v)
        data = formulation.data()
        contact_Point = np.ones((3, 4)) * conf.lz
        contact_Point[0, :] = [-conf.lxn, -conf.lxn, conf.lxp, conf.lxp]
        contact_Point[1, :] = [-conf.lyn, conf.lyp, -conf.lyn, conf.lyp]

        contactRF = tsid.Contact6d("contact_rfoot", robot, conf.rf_frame_name,
                                   contact_Point, conf.contactNormal, conf.mu,
                                   conf.fMin, conf.fMax)
        contactRF.setKp(conf.kp_contact * np.ones(6))
        contactRF.setKd(2.0 * np.sqrt(conf.kp_contact) * np.ones(6))
        self.RF = robot.model().getFrameId(conf.rf_frame_name)
        H_rf_ref = robot.framePosition(data, self.RF)

        # modify initial robot configuration so that foot is on the ground (z=0)
        q[2] -= H_rf_ref.translation[2] - conf.lz
        formulation.computeProblemData(0.0, q, v)
        data = formulation.data()
        H_rf_ref = robot.framePosition(data, self.RF)
        contactRF.setReference(H_rf_ref)
        if conf.w_contact >= 0.0:
            formulation.addRigidContact(contactRF, conf.w_forceRef,
                                        conf.w_contact, 1)
        else:
            formulation.addRigidContact(contactRF, conf.w_forceRef)

        contactLF = tsid.Contact6d("contact_lfoot", robot, conf.lf_frame_name,
                                   contact_Point, conf.contactNormal, conf.mu,
                                   conf.fMin, conf.fMax)
        contactLF.setKp(conf.kp_contact * np.ones(6))
        contactLF.setKd(2.0 * np.sqrt(conf.kp_contact) * np.ones(6))
        self.LF = robot.model().getFrameId(conf.lf_frame_name)
        H_lf_ref = robot.framePosition(data, self.LF)
        contactLF.setReference(H_lf_ref)
        if conf.w_contact >= 0.0:
            formulation.addRigidContact(contactLF, conf.w_forceRef,
                                        conf.w_contact, 1)
        else:
            formulation.addRigidContact(contactLF, conf.w_forceRef)

        comTask = tsid.TaskComEquality("task-com", robot)
        comTask.setKp(conf.kp_com * np.ones(3))
        comTask.setKd(2.0 * np.sqrt(conf.kp_com) * np.ones(3))
        formulation.addMotionTask(comTask, conf.w_com, 1, 0.0)

        postureTask = tsid.TaskJointPosture("task-posture", robot)
        postureTask.setKp(conf.kp_posture * conf.gain_vector)
        postureTask.setKd(2.0 * np.sqrt(conf.kp_posture * conf.gain_vector))
        postureTask.mask(conf.masks_posture)
        formulation.addMotionTask(postureTask, conf.w_posture, 1, 0.0)

        orientationRootTask = tsid.TaskSE3Equality("task-orientation-root",
                                                   robot, 'root_joint')
        mask = np.ones(6)
        mask[0:3] = 0
        orientationRootTask.setMask(mask)
        orientationRootTask.setKp(conf.kp_rootOrientation * mask)
        orientationRootTask.setKd(2.0 *
                                  np.sqrt(conf.kp_rootOrientation * mask))
        trajoriRoot = tsid.TrajectorySE3Constant("traj-ori",
                                                 pin.SE3().Identity())
        formulation.addMotionTask(orientationRootTask, conf.w_rootOrientation,
                                  1, 0.0)

        self.amTask = tsid.TaskAMEquality("task-am", robot)
        self.amTask.setKp(conf.kp_am * np.array([1., 1., 0.]))
        self.amTask.setKd(2.0 * np.sqrt(conf.kp_am * np.array([1., 1., 0.])))
        formulation.addMotionTask(self.amTask, conf.w_am, 1, 0.)
        self.sampleAM = tsid.TrajectorySample(3)
        self.amTask.setReference(self.sampleAM)

        self.leftFootTask = tsid.TaskSE3Equality("task-left-foot", self.robot,
                                                 self.conf.lf_frame_name)
        self.leftFootTask.setKp(self.conf.kp_foot * np.ones(6))
        self.leftFootTask.setKd(2.0 * np.sqrt(self.conf.kp_foot) * np.ones(6))
        self.trajLF = tsid.TrajectorySE3Constant("traj-left-foot", H_lf_ref)
        formulation.addMotionTask(self.leftFootTask, self.conf.w_foot, 1, 0.0)

        self.rightFootTask = tsid.TaskSE3Equality("task-right-foot",
                                                  self.robot,
                                                  self.conf.rf_frame_name)
        self.rightFootTask.setKp(self.conf.kp_foot * np.ones(6))
        self.rightFootTask.setKd(2.0 * np.sqrt(self.conf.kp_foot) * np.ones(6))
        self.trajRF = tsid.TrajectorySE3Constant("traj-right-foot", H_rf_ref)
        formulation.addMotionTask(self.rightFootTask, self.conf.w_foot, 1, 0.0)

        if self.t_limit:
            self.tau_max = conf.tau_max_scaling * robot.model(
            ).effortLimit[-robot.na:]
            self.tau_min = -self.tau_max
            actuationBoundsTask = tsid.TaskActuationBounds(
                "task-actuation-bounds", robot)
            actuationBoundsTask.setBounds(self.tau_min, self.tau_max)
            if conf.w_torque_bounds > 0.0:
                print("torque_limit")
                formulation.addActuationTask(actuationBoundsTask,
                                             conf.w_torque_bounds, 0, 0.0)

        if self.j_limit:
            jointBoundsTask = tsid.TaskJointBounds("task-joint-bounds", robot,
                                                   conf.dt)
            self.v_max = conf.v_max_scaling * robot.model(
            ).velocityLimit[-robot.na:]
            self.v_min = -self.v_max
            jointBoundsTask.setVelocityBounds(self.v_min, self.v_max)
            if conf.w_joint_bounds > 0.0:
                print("joint_limit")
                formulation.addMotionTask(jointBoundsTask, conf.w_joint_bounds,
                                          0, 0.0)

        com_ref = robot.com(data)
        self.trajCom = tsid.TrajectoryEuclidianConstant("traj_com", com_ref)
        self.sample_com = self.trajCom.computeNext()

        q_ref = q[7:]
        q_ref[:12] = np.array([
            0., 0., -0.4114, 0.8594, -0.448, -0.0017, 0., 0., -0.4114, 0.8594,
            -0.448, -0.0017
        ])
        self.trajPosture = tsid.TrajectoryEuclidianConstant(
            "traj_joint", q_ref)
        postureTask.setReference(self.trajPosture.computeNext())

        orientationRootTask.setReference(trajoriRoot.computeNext())

        self.sampleLF = self.trajLF.computeNext()
        self.sample_LF_pos = self.sampleLF.pos()
        self.sample_LF_vel = self.sampleLF.vel()
        self.sample_LF_acc = self.sampleLF.acc()

        self.sampleRF = self.trajRF.computeNext()
        self.sample_RF_pos = self.sampleRF.pos()
        self.sample_RF_vel = self.sampleRF.vel()
        self.sample_RF_acc = self.sampleRF.acc()

        self.solver = tsid.SolverHQuadProgFast("qp solver")
        self.solver.resize(formulation.nVar, formulation.nEq, formulation.nIn)

        self.comTask = comTask
        self.postureTask = postureTask
        self.contactRF = contactRF
        self.contactLF = contactLF
        if self.t_limit:
            self.actuationBoundsTask = actuationBoundsTask
        if self.j_limit:
            self.jointBoundsTask = jointBoundsTask
        self.orientationRootTask = orientationRootTask
        self.formulation = formulation
        self.q = q
        self.v = v

        self.contact_LF_active = True
        self.contact_RF_active = True

        if viewer:
            self.robot_display = pin.RobotWrapper.BuildFromURDF(
                conf.urdf, [conf.path], pin.JointModelFreeFlyer())
            if viewer == pin.visualize.GepettoVisualizer:
                import gepetto.corbaserver
                launched = subprocess.getstatusoutput(
                    "ps aux |grep 'gepetto-gui'|grep -v 'grep'|wc -l")
                if int(launched[1]) == 0:
                    os.system('gepetto-gui &')
                time.sleep(1)
                self.viz = viewer(self.robot_display.model,
                                  self.robot_display.collision_model,
                                  self.robot_display.visual_model)
                self.viz.initViewer(loadModel=True)
                self.viz.displayCollisions(False)
                self.viz.displayVisuals(True)
                self.viz.display(q)

                self.gui = self.viz.viewer.gui
                # self.gui.setCameraTransform(0, conf.CAMERA_TRANSFORM)
                # self.gui.addFloor('world/floor')
                self.gui.setLightingMode('world/floor', 'OFF')
            elif viewer == pin.visualize.MeshcatVisualizer:
                self.viz = viewer(self.robot_display.model,
                                  self.robot_display.collision_model,
                                  self.robot_display.visual_model)
                self.viz.initViewer(loadModel=True)
                self.viz.display(q)
Exemple #15
0
    def __init__(self):

        ########################################################################
        #             Definition of the Model and TSID problem                 #
        ########################################################################

        # set the path
        path = "/opt/openrobots/share/example-robot-data/robots/solo_description/robots"
        urdf = path + "/solo12.urdf"
        srdf = "/opt/openrobots/share/example-robot-data/robots/solo_description/srdf/solo.srdf"
        vec = pin.StdVec_StdString()
        vec.extend(item for item in path)

        # get the solo12 wrapper
        self.solo12_wrapper = tsid.RobotWrapper(urdf, vec,
                                                pin.JointModelFreeFlyer(),
                                                False)
        self.solo12_model = self.solo12_wrapper.model()
        pin.loadReferenceConfigurations(self.solo12_model, srdf, False)

        # problem formulation
        self.formulation = tsid.InverseDynamicsFormulationAccForce(
            "tsid", self.solo12_wrapper, False)
        self.q0 = self.solo12_model.referenceConfigurations[
            "straight_standing"]
        self.v0 = np.zeros(self.solo12_model.nv)
        self.formulation.computeProblemData(0.0, self.q0, self.v0)
        self.data = self.formulation.data()

        # get frame ID
        self.ID_FL = self.solo12_model.getFrameId("FL_FOOT")
        self.ID_FR = self.solo12_model.getFrameId("FR_FOOT")
        self.ID_HL = self.solo12_model.getFrameId("HL_FOOT")
        self.ID_HR = self.solo12_model.getFrameId("HR_FOOT")
        self.ID_BASE = self.solo12_model.getFrameId("base_link")

        ############
        # COM TASK #
        ############
        self.kp_com = 50
        self.w_com = 3
        self.comTask = tsid.TaskComEquality("task-com", self.solo12_wrapper)
        self.comTask.setKp(self.kp_com * np.ones(3).T)
        self.comTask.setKd(2 * np.sqrt(self.kp_com) * np.ones(3).T)
        self.formulation.addMotionTask(self.comTask, self.w_com, 1, 0.0)
        self.com_ref = self.solo12_wrapper.com(self.data)
        self.trajCom = tsid.TrajectoryEuclidianConstant(
            "traj-com", self.com_ref)

        #################
        # SE3 BASE TASK #
        #################
        self.kp_trunk = 50
        self.w_trunk = 1
        self.trunkTask = tsid.TaskSE3Equality("task-trunk",
                                              self.solo12_wrapper, 'base_link')
        mask = np.matrix([1.0, 1.0, 0.0, 1.0, 1.0, 1.0]).T
        self.trunkTask.setKp(
            np.matrix([
                self.kp_trunk, self.kp_trunk, 0.0, self.kp_trunk,
                self.kp_trunk, self.kp_trunk
            ]).T)
        self.trunkTask.setKd(
            np.matrix([
                2.0 * np.sqrt(self.kp_trunk), 2.0 * np.sqrt(self.kp_trunk),
                0.0, 2.0 * np.sqrt(self.kp_trunk),
                2.0 * np.sqrt(self.kp_trunk), 2.0 * np.sqrt(self.kp_trunk)
            ]).T)
        self.trunkTask.useLocalFrame(False)
        self.trunkTask.setMask(mask)
        self.formulation.addMotionTask(self.trunkTask, self.w_trunk, 1, 0.0)
        self.trunk_ref = self.solo12_wrapper.framePosition(
            self.data, self.ID_BASE)
        self.trajTrunk = tsid.TrajectorySE3Constant("traj_base_link",
                                                    self.trunk_ref)

        ##################
        # CONTACT FORCES #
        ##################

        self.contactNormal = np.array([0, 0, 1])
        self.kp_contact = 30
        self.mu = 0.5

        self.contactFL = tsid.ContactPoint("contactFL", self.solo12_wrapper,
                                           "FL_FOOT", self.contactNormal,
                                           self.mu, 1, 1000)
        self.contactFL.setKp(self.kp_contact * np.ones(6).T)
        self.contactFL.setKd(2 * np.sqrt(self.kp_contact) * np.ones(6).T)
        self.fl_ref = self.solo12_wrapper.framePosition(self.data, self.ID_FL)
        self.contactFL.setReference(self.fl_ref)
        self.formulation.addRigidContact(self.contactFL, 10)

        self.contactFR = tsid.ContactPoint("contactFR", self.solo12_wrapper,
                                           "FR_FOOT", self.contactNormal,
                                           self.mu, 1, 1000)
        self.contactFR.setKp(self.kp_contact * np.ones(6).T)
        self.contactFR.setKd(2 * np.sqrt(self.kp_contact) * np.ones(6).T)
        self.fr_ref = self.solo12_wrapper.framePosition(self.data, self.ID_FR)
        self.contactFR.setReference(self.fr_ref)
        self.formulation.addRigidContact(self.contactFR, 10)

        self.contactHR = tsid.ContactPoint("contactHR", self.solo12_wrapper,
                                           "HR_FOOT", self.contactNormal,
                                           self.mu, 1, 1000)
        self.contactHR.setKp(self.kp_contact * np.ones(6).T)
        self.contactHR.setKd(2 * np.sqrt(self.kp_contact) * np.ones(6).T)
        self.hr_ref = self.solo12_wrapper.framePosition(self.data, self.ID_HR)
        self.contactHR.setReference(self.hr_ref)
        self.formulation.addRigidContact(self.contactHR, 10)

        self.contactHL = tsid.ContactPoint("contactHL", self.solo12_wrapper,
                                           "HL_FOOT", self.contactNormal,
                                           self.mu, 1, 1000)
        self.contactHL.setKp(self.kp_contact * np.ones(6).T)
        self.contactHL.setKd(2 * np.sqrt(self.kp_contact) * np.ones(6).T)
        self.hl_ref = self.solo12_wrapper.framePosition(self.data, self.ID_HL)
        self.contactHL.setReference(self.hl_ref)
        self.formulation.addRigidContact(self.contactHL, 10)

        ##########
        # SOLVER #
        ##########

        self.solver = tsid.SolverHQuadProgFast("qp solver")
        self.solver.resize(self.formulation.nVar, self.formulation.nEq,
                           self.formulation.nIn)
Exemple #16
0
    def __init__(self, conf, viewer=True):
        self.conf = conf
        self.contact_frame_names = conf.contact_frame_names
        print('Robot files:')
        print(conf.urdf)
        print(conf.srdf)
        vector = pin.StdVec_StdString()
        vector.extend(item for item in conf.path)
        self.robot = tsid.RobotWrapper(conf.urdf, vector,
                                       pin.JointModelFreeFlyer(), False)
        robot = self.robot
        self.model = robot.model()

        pin.loadReferenceConfigurations(self.model, conf.srdf, False)
        self.q0 = q = self.model.referenceConfigurations[
            conf.reference_config_q_name]
        v = np.zeros(robot.nv)

        assert (all([
            self.model.existFrame(frame_name)
            for frame_name in self.contact_frame_names
        ]))

        invdyn = tsid.InverseDynamicsFormulationAccForce('tsid', robot, False)
        invdyn.computeProblemData(0.0, q, v)
        data = invdyn.data()

        robot.computeAllTerms(data, q, v)

        #####################################
        # define contact and associated tasks
        #####################################

        self.contact_frame_ids = [
            self.model.getFrameId(frame_name)
            for frame_name in self.contact_frame_names
        ]
        self.nc = len(self.contact_frame_ids)  # contact number

        self.contacts = self.nc * [None]
        self.tasks_tracking_foot = self.nc * [None]
        self.traj_feet = self.nc * [None]
        self.sample_feet = self.nc * [None]
        mask = np.ones(6)
        if not conf.contact6d:
            mask[3:] = 0
        for i_foot, (frame_name, fid) in enumerate(
                zip(self.contact_frame_names, self.contact_frame_ids)):
            Hf_ref = robot.framePosition(data, fid)

            if conf.contact6d:
                # DEFINE FOOT CONTACT POINTS LOCATION WRT LOCAL FOOT FRAMES
                # contact_points = np.ones((3,4)) * conf.lz
                contact_points = -np.ones((3, 4)) * conf.lz
                contact_points[0, :] = [
                    -conf.lxn, -conf.lxn, conf.lxp, conf.lxp
                ]
                contact_points[1, :] = [
                    -conf.lyn, conf.lyp, -conf.lyn, conf.lyp
                ]

                # RIGID CONTACT RIGHT FOOT
                self.contacts[i_foot] = tsid.Contact6d(
                    'contact_{}'.format(frame_name), robot, frame_name,
                    contact_points, conf.contactNormal, conf.mu, conf.fMin,
                    conf.fMax)
                self.contacts[i_foot].setKp(conf.kp_contact * np.ones(6).T)
                self.contacts[i_foot].setKd(2.0 * np.sqrt(conf.kp_contact) *
                                            np.ones(6).T)
                self.contacts[i_foot].setReference(Hf_ref)
                invdyn.addRigidContact(self.contacts[i_foot], conf.w_forceRef)

            else:
                # RIGID POINT CONTACTS
                self.contacts[i_foot] = tsid.ContactPoint(
                    'contact_{}'.format(frame_name), robot, frame_name,
                    conf.contactNormal, conf.mu, conf.fMin, conf.fMax)
                self.contacts[i_foot].setKp(conf.kp_contact * np.ones(3))
                self.contacts[i_foot].setKd(2.0 * np.sqrt(conf.kp_contact) *
                                            np.ones(3))
                self.contacts[i_foot].setReference(Hf_ref)
                self.contacts[i_foot].useLocalFrame(conf.useLocalFrame)
                invdyn.addRigidContact(self.contacts[i_foot], conf.w_forceRef)

            # FEET TRACKING TASKS
            self.tasks_tracking_foot[i_foot] = tsid.TaskSE3Equality(
                'task-foot{}'.format(i_foot), self.robot,
                self.contact_frame_names[i_foot])
            self.tasks_tracking_foot[i_foot].setKp(conf.kp_foot * mask)
            self.tasks_tracking_foot[i_foot].setKd(
                2.0 * np.sqrt(conf.kp_foot) * mask)
            self.tasks_tracking_foot[i_foot].setMask(mask)
            self.tasks_tracking_foot[i_foot].useLocalFrame(False)
            invdyn.addMotionTask(self.tasks_tracking_foot[i_foot], conf.w_foot,
                                 conf.priority_foot, 0.0)

            self.traj_feet[i_foot] = tsid.TrajectorySE3Constant(
                'traj-foot{}'.format(i_foot), Hf_ref)
            self.sample_feet[i_foot] = self.traj_feet[i_foot].computeNext()

        #############
        # Other tasks
        #############

        # COM TASK
        self.comTask = tsid.TaskComEquality('task-com', robot)
        self.comTask.setKp(conf.kp_com * np.ones(3))
        self.comTask.setKd(2.0 * np.sqrt(conf.kp_com) * np.ones(3))
        invdyn.addMotionTask(self.comTask, conf.w_com, conf.priority_com, 1.0)

        # POSTURE TASK
        self.postureTask = tsid.TaskJointPosture('task-posture', robot)
        self.postureTask.setKp(conf.kp_posture * np.ones(robot.nv - 6))
        self.postureTask.setKd(2.0 * np.sqrt(conf.kp_posture) *
                               np.ones(robot.nv - 6))
        invdyn.addMotionTask(self.postureTask, conf.w_posture,
                             conf.priority_posture, 0.0)

        # TRUNK Task
        self.trunkTask = tsid.TaskSE3Equality("keepTrunk", robot, 'root_joint')
        self.trunkTask.setKp(conf.kp_trunk * np.ones(6))
        self.trunkTask.setKd(2.0 * np.sqrt(conf.kp_trunk) * np.ones(6))
        # Add a Mask to the task which will select the vector dimensions on which the task will act.
        # In this case the trunk configuration is a vector 6d (position and orientation -> SE3)
        # Here we set a mask = [0 0 0 1 1 1] so the task on the trunk will act on the orientation of the robot
        mask = np.ones(6)
        mask[:3] = 0.
        self.trunkTask.useLocalFrame(False)
        self.trunkTask.setMask(mask)
        invdyn.addMotionTask(self.trunkTask, conf.w_trunk, conf.priority_trunk,
                             0.0)

        # TORQUE BOUND TASK
        # self.tau_max = conf.tau_max_scaling*robot.model().effortLimit[-robot.na:]
        # self.tau_min = -self.tau_max
        # actuationBoundsTask = tsid.TaskActuationBounds('task-actuation-bounds', robot)
        # actuationBoundsTask.setBounds(self.tau_min, self.tau_max)
        # if(conf.w_torque_bounds>0.0):
        #     invdyn.addActuationTask(actuationBoundsTask, conf.w_torque_bounds, conf.priority_torque_bounds, 0.0)

        # JOINT BOUND TASK
        # jointBoundsTask = tsid.TaskJointBounds('task-joint-bounds', robot, conf.dt)
        # self.v_max = conf.v_max_scaling * robot.model().velocityLimit[-robot.na:]
        # self.v_min = -self.v_max
        # jointBoundsTask.setVelocityBounds(self.v_min, self.v_max)
        # if(conf.w_joint_bounds>0.0):
        #     invdyn.addMotionTask(jointBoundsTask, conf.w_joint_bounds, conf.priority_joint_bounds, 0.0)

        ##################################
        # Init task reference trajectories
        ##################################

        self.sample_feet = self.nc * [None]
        self.sample_feet_pos = self.nc * [None]
        self.sample_feet_vel = self.nc * [None]
        self.sample_feet_acc = self.nc * [None]
        for i_foot, traj in enumerate(self.traj_feet):
            self.sample_feet[i_foot] = traj.computeNext()
            self.sample_feet_pos[i_foot] = self.sample_feet[i_foot].pos()
            self.sample_feet_vel[i_foot] = self.sample_feet[i_foot].vel()
            self.sample_feet_acc[i_foot] = self.sample_feet[i_foot].acc()

        com_ref = robot.com(data)
        self.trajCom = tsid.TrajectoryEuclidianConstant('traj_com', com_ref)
        self.sample_com = self.trajCom.computeNext()

        q_ref = q[7:]
        self.trajPosture = tsid.TrajectoryEuclidianConstant(
            'traj_joint', q_ref)
        self.postureTask.setReference(self.trajPosture.computeNext())

        self.trunk_link_id = self.model.getFrameId('base_link')
        R_trunk_init = robot.framePosition(data, self.trunk_link_id).rotation
        self.trunk_ref = self.robot.framePosition(data, self.trunk_link_id)
        self.trajTrunk = tsid.TrajectorySE3Constant("traj_base_link",
                                                    self.trunk_ref)
        self.sample_trunk = self.trajTrunk.computeNext()
        pos_trunk = np.hstack([np.zeros(3), R_trunk_init.flatten()])
        self.sample_trunk.pos()
        self.sample_trunk.vel(np.zeros(6))
        self.sample_trunk.acc(np.zeros(6))
        self.trunkTask.setReference(self.sample_trunk)

        self.solver = tsid.SolverHQuadProgFast('qp solver')
        self.solver.resize(invdyn.nVar, invdyn.nEq, invdyn.nIn)

        self.invdyn = invdyn
        self.q = q
        self.v = v

        self.active_contacts = self.nc * [True]

        # for gepetto viewer
        if (viewer):
            self.robot_display = pin.RobotWrapper.BuildFromURDF(
                conf.urdf, [
                    conf.path,
                ], pin.JointModelFreeFlyer())
            l = subprocess.getstatusoutput(
                "ps aux |grep 'gepetto-gui'|grep -v 'grep'|wc -l")
            if int(l[1]) == 0:
                os.system('gepetto-gui &')
            time.sleep(1)
            gepetto.corbaserver.Client()
            self.robot_display.initViewer(loadModel=True)
            self.robot_display.displayCollisions(False)
            self.robot_display.displayVisuals(True)
            self.robot_display.display(q)

            self.gui = self.robot_display.viewer.gui
            self.gui.setCameraTransform('python-pinocchio',
                                        conf.CAMERA_TRANSFORM)
            self.gui.addFloor('world/floor')
            self.gui.setLightingMode('world/floor', 'OFF')

            self.gui.addSphere('world/com', conf.SPHERE_RADIUS,
                               conf.COM_SPHERE_COLOR)
            self.gui.addSphere('world/com_ref', conf.REF_SPHERE_RADIUS,
                               conf.COM_REF_SPHERE_COLOR)
            for frame_name in self.contact_frame_names:
                self.gui.addSphere('world/{}'.format(frame_name),
                                   conf.SPHERE_RADIUS, conf.F_SPHERE_COLOR)
                self.gui.addSphere('world/{}_ref'.format(frame_name),
                                   conf.REF_SPHERE_RADIUS,
                                   conf.F_REF_SPHERE_COLOR)
Exemple #17
0
    def __init__(self, conf, viewer=True):
        self.conf = conf
        vector = se3.StdVec_StdString()
        vector.extend(item for item in conf.path)
        self.robot = tsid.RobotWrapper(conf.urdf, vector,
                                       se3.JointModelFreeFlyer(), False)
        robot = self.robot
        self.model = robot.model()
        pin.loadReferenceConfigurations(self.model, conf.srdf, False)
        self.q0 = q = self.model.referenceConfigurations["half_sitting"]
        v = np.zeros(robot.nv)

        assert self.model.existFrame(conf.rf_frame_name)
        assert self.model.existFrame(conf.lf_frame_name)

        formulation = tsid.InverseDynamicsFormulationAccForce(
            "tsid", robot, False)
        formulation.computeProblemData(0.0, q, v)
        data = formulation.data()
        contact_Point = np.ones((3, 4)) * conf.lz
        contact_Point[0, :] = [-conf.lxn, -conf.lxn, conf.lxp, conf.lxp]
        contact_Point[1, :] = [-conf.lyn, conf.lyp, -conf.lyn, conf.lyp]

        contactRF = tsid.Contact6d("contact_rfoot", robot, conf.rf_frame_name,
                                   contact_Point, conf.contactNormal, conf.mu,
                                   conf.fMin, conf.fMax)
        contactRF.setKp(conf.kp_contact * np.ones(6))
        contactRF.setKd(2.0 * np.sqrt(conf.kp_contact) * np.ones(6))
        self.RF = robot.model().getJointId(conf.rf_frame_name)
        H_rf_ref = robot.position(data, self.RF)

        # modify initial robot configuration so that foot is on the ground (z=0)
        q[2] -= H_rf_ref.translation[2] - conf.lz
        formulation.computeProblemData(0.0, q, v)
        data = formulation.data()
        H_rf_ref = robot.position(data, self.RF)
        contactRF.setReference(H_rf_ref)
        formulation.addRigidContact(contactRF, conf.w_forceRef)

        contactLF = tsid.Contact6d("contact_lfoot", robot, conf.lf_frame_name,
                                   contact_Point, conf.contactNormal, conf.mu,
                                   conf.fMin, conf.fMax)
        contactLF.setKp(conf.kp_contact * np.ones(6))
        contactLF.setKd(2.0 * np.sqrt(conf.kp_contact) * np.ones(6))
        self.LF = robot.model().getJointId(conf.lf_frame_name)
        H_lf_ref = robot.position(data, self.LF)
        contactLF.setReference(H_lf_ref)
        formulation.addRigidContact(contactLF, conf.w_forceRef)

        comTask = tsid.TaskComEquality("task-com", robot)
        comTask.setKp(conf.kp_com * np.ones(3))
        comTask.setKd(2.0 * np.sqrt(conf.kp_com) * np.ones(3))
        formulation.addMotionTask(comTask, conf.w_com, 1, 0.0)

        postureTask = tsid.TaskJointPosture("task-posture", robot)
        postureTask.setKp(conf.kp_posture * np.ones(robot.nv - 6))
        postureTask.setKd(2.0 * np.sqrt(conf.kp_posture) *
                          np.ones(robot.nv - 6))
        formulation.addMotionTask(postureTask, conf.w_posture, 1, 0.0)

        self.leftFootTask = tsid.TaskSE3Equality("task-left-foot", self.robot,
                                                 self.conf.lf_frame_name)
        self.leftFootTask.setKp(self.conf.kp_foot * np.ones(6))
        self.leftFootTask.setKd(2.0 * np.sqrt(self.conf.kp_foot) * np.ones(6))
        self.trajLF = tsid.TrajectorySE3Constant("traj-left-foot", H_lf_ref)
        formulation.addMotionTask(self.leftFootTask, self.conf.w_foot, 1, 0.0)

        self.rightFootTask = tsid.TaskSE3Equality("task-right-foot",
                                                  self.robot,
                                                  self.conf.rf_frame_name)
        self.rightFootTask.setKp(self.conf.kp_foot * np.ones(6))
        self.rightFootTask.setKd(2.0 * np.sqrt(self.conf.kp_foot) * np.ones(6))
        self.trajRF = tsid.TrajectorySE3Constant("traj-right-foot", H_rf_ref)
        formulation.addMotionTask(self.rightFootTask, self.conf.w_foot, 1, 0.0)

        self.tau_max = conf.tau_max_scaling * robot.model(
        ).effortLimit[-robot.na:]
        self.tau_min = -self.tau_max
        actuationBoundsTask = tsid.TaskActuationBounds("task-actuation-bounds",
                                                       robot)
        actuationBoundsTask.setBounds(self.tau_min, self.tau_max)
        if (conf.w_torque_bounds > 0.0):
            formulation.addActuationTask(actuationBoundsTask,
                                         conf.w_torque_bounds, 0, 0.0)

        jointBoundsTask = tsid.TaskJointBounds("task-joint-bounds", robot,
                                               conf.dt)
        self.v_max = conf.v_max_scaling * robot.model(
        ).velocityLimit[-robot.na:]
        self.v_min = -self.v_max
        jointBoundsTask.setVelocityBounds(self.v_min, self.v_max)
        if (conf.w_joint_bounds > 0.0):
            formulation.addMotionTask(jointBoundsTask, conf.w_joint_bounds, 0,
                                      0.0)

        com_ref = robot.com(data)
        self.trajCom = tsid.TrajectoryEuclidianConstant("traj_com", com_ref)
        self.sample_com = self.trajCom.computeNext()

        q_ref = q[7:]
        self.trajPosture = tsid.TrajectoryEuclidianConstant(
            "traj_joint", q_ref)
        postureTask.setReference(self.trajPosture.computeNext())

        self.sampleLF = self.trajLF.computeNext()
        self.sample_LF_pos = self.sampleLF.pos()
        self.sample_LF_vel = self.sampleLF.vel()
        self.sample_LF_acc = self.sampleLF.acc()

        self.sampleRF = self.trajRF.computeNext()
        self.sample_RF_pos = self.sampleRF.pos()
        self.sample_RF_vel = self.sampleRF.vel()
        self.sample_RF_acc = self.sampleRF.acc()

        self.solver = tsid.SolverHQuadProgFast("qp solver")
        self.solver.resize(formulation.nVar, formulation.nEq, formulation.nIn)

        self.comTask = comTask
        self.postureTask = postureTask
        self.contactRF = contactRF
        self.contactLF = contactLF
        self.actuationBoundsTask = actuationBoundsTask
        self.jointBoundsTask = jointBoundsTask
        self.formulation = formulation
        self.q = q
        self.v = v

        self.contact_LF_active = True
        self.contact_RF_active = True

        # for gepetto viewer
        if (viewer):
            self.robot_display = se3.RobotWrapper.BuildFromURDF(
                conf.urdf, [
                    conf.path,
                ], se3.JointModelFreeFlyer())
            l = subprocess.getstatusoutput(
                "ps aux |grep 'gepetto-gui'|grep -v 'grep'|wc -l")
            if int(l[1]) == 0:
                os.system('gepetto-gui &')
            time.sleep(1)
            gepetto.corbaserver.Client()
            self.robot_display.initViewer(loadModel=True)
            self.robot_display.displayCollisions(False)
            self.robot_display.displayVisuals(True)
            self.robot_display.display(q)

            self.gui = self.robot_display.viewer.gui
            #            self.gui.setCameraTransform(0, conf.CAMERA_TRANSFORM)
            self.gui.addFloor('world/floor')
            self.gui.setLightingMode('world/floor', 'OFF')
Exemple #18
0
postureTask = tsid.TaskJointPosture("task-posture", robot)
postureTask.setKp(kp_posture * np.matrix(np.ones(robot.nv-6)).transpose())
postureTask.setKd(2.0 * np.sqrt(kp_posture) * np.matrix(np.ones(robot.nv-6)).transpose())
invdyn.addMotionTask(postureTask, w_posture, 1, 0.0)

rightFootTask = tsid.TaskSE3Equality("task-right-foot", robot, rf_frame_name)
rightFootTask.setKp(kp_RF * np.matrix(np.ones(6)).transpose())
rightFootTask.setKd(2.0 * np.sqrt(kp_com) * np.matrix(np.ones(6)).transpose())
invdyn.addMotionTask(rightFootTask, w_RF, 1, 0.0)

H_rf_ref.translation += np.matrix([0., 0., DELTA_FOOT_Z]).T
rightFootTraj = tsid.TrajectorySE3Constant("traj-right-foot", H_rf_ref)

com_ref = robot.com(data)
com_ref[1] += DELTA_COM_Y
trajCom = tsid.TrajectoryEuclidianConstant("traj_com", com_ref)

q_ref = q[7:]
trajPosture = tsid.TrajectoryEuclidianConstant("traj_joint", q_ref)

solver = tsid.SolverHQuadProg("qp solver")
solver.resize(invdyn.nVar, invdyn.nEq, invdyn.nIn)

for i in range(0, N_SIMULATION):
    time_start = time.time()
    
    if i == REMOVE_CONTACT_N:
        print "\nTime %.3f Start breaking contact %s\n"%(t, contactRF.name)
        invdyn.removeRigidContact(contactRF.name, CONTACT_TRANSITION_TIME)

    sampleCom = trajCom.computeNext()
Exemple #19
0
    def __init__(self,
                 conf,
                 dt,
                 robot,
                 com_offset,
                 com_frequency,
                 com_amplitude,
                 q0=None,
                 viewer=True):
        self.conf = conf
        self.dt = dt
        self.robot = tsid.RobotWrapper(robot.model, False)

        if q0 is None:
            q = robot.model.neutralConfiguration
        else:
            q = np.copy(q0)
#        q = se3.getNeutralConfiguration(robot.model(), conf.srdf, False)
#        q = robot.model().referenceConfigurations["half_sitting"]
        v = np.zeros(robot.nv)

        # for gepetto viewer
        if (viewer):
            # se3.RobotWrapper.BuildFromURDF(conf.urdf, [conf.path, ], se3.JointModelFreeFlyer())
            self.robot_display = robot
            prompt = subprocess.getstatusoutput(
                "ps aux |grep 'gepetto-gui'|grep -v 'grep'|wc -l")
            if int(prompt[1]) == 0:
                os.system('gepetto-gui &')
            time.sleep(1)
            gepetto.corbaserver.Client()
            self.robot_display.initViewer(loadModel=True)
            self.robot_display.displayCollisions(False)
            self.robot_display.displayVisuals(True)
            self.robot_display.display(q)
            self.gui = self.robot_display.viewer.gui
            self.gui.setCameraTransform('gepetto', conf.CAMERA_TRANSFORM)

        robot = self.robot
        self.model = robot.model()

        #        assert [robot.model().existFrame(name) for name in conf.contact_frames]

        formulation = tsid.InverseDynamicsFormulationAccForce(
            "tsid", robot, False)
        formulation.computeProblemData(0.0, q, v)
        data = formulation.data()

        contacts = len(conf.contact_frames) * [None]
        self.contact_ids = {}
        for i, name in enumerate(conf.contact_frames):
            contacts[i] = tsid.ContactPoint(name, robot, name,
                                            conf.contact_normal, conf.mu,
                                            conf.fMin, conf.fMax)
            contacts[i].setKp(conf.kp_contact * np.ones(3))
            contacts[i].setKd(2.0 * np.sqrt(conf.kp_contact) * np.ones(3))
            self.contact_ids[name] = robot.model().getFrameId(name)
            H_ref = robot.framePosition(data, robot.model().getFrameId(name))
            contacts[i].setReference(H_ref)
            contacts[i].useLocalFrame(False)
            formulation.addRigidContact(contacts[i], conf.w_forceRef, 1.0, 1)

        # modify initial robot configuration so that foot is on the ground (z=0)
#        q[2] -= H_rf_ref.translation[2,0] # 0.84    # fix this
#        formulation.computeProblemData(0.0, q, v)
#        data = formulation.data()
#        H_rf_ref = robot.position(data, self.RF)
#        contactRF.setReference(H_rf_ref)
#        formulation.addRigidContact(contactRF, conf.w_forceRef)

        comTask = tsid.TaskComEquality("task-com", robot)
        comTask.setKp(conf.kp_com * np.ones(3))
        comTask.setKd(2.0 * np.sqrt(conf.kp_com) * np.ones(3))
        formulation.addMotionTask(comTask, conf.w_com, 1, 0.0)

        postureTask = tsid.TaskJointPosture("task-posture", robot)
        postureTask.setKp(conf.kp_posture * np.ones(robot.nv - 6))
        postureTask.setKd(2.0 * np.sqrt(conf.kp_posture) *
                          np.ones(robot.nv - 6))
        formulation.addMotionTask(postureTask, conf.w_posture, 1, 0.0)

        #        self.leftFootTask = tsid.TaskSE3Equality("task-left-foot", self.robot, self.conf.lf_frame_name)
        #        self.leftFootTask.setKp(self.conf.kp_foot * np.array(np.ones(6)).T)
        #        self.leftFootTask.setKd(2.0 * np.sqrt(self.conf.kp_foot) * np.array(np.ones(6)).T)
        #        self.trajLF = tsid.TrajectorySE3Constant("traj-left-foot", H_lf_ref)
        #
        #        self.rightFootTask = tsid.TaskSE3Equality("task-right-foot", self.robot, self.conf.rf_frame_name)
        #        self.rightFootTask.setKp(self.conf.kp_foot * np.array(np.ones(6)).T)
        #        self.rightFootTask.setKd(2.0 * np.sqrt(self.conf.kp_foot) * np.array(np.ones(6)).T)
        #        self.trajRF = tsid.TrajectorySE3Constant("traj-right-foot", H_rf_ref)
        #
        com_ref = robot.com(data)
        trajCom = tsid.TrajectoryEuclidianConstant("traj_com", com_ref)

        q_ref = q[7:]
        trajPosture = tsid.TrajectoryEuclidianConstant("traj_joint", q_ref)

        comTask.setReference(trajCom.computeNext())
        postureTask.setReference(trajPosture.computeNext())

        solver = tsid.SolverHQuadProgFast("qp solver")
        solver.resize(formulation.nVar, formulation.nEq, formulation.nIn)

        self.trajCom = trajCom
        self.trajPosture = trajPosture
        self.comTask = comTask
        self.postureTask = postureTask
        self.contacts = contacts
        self.formulation = formulation
        self.solver = solver
        self.q = q
        self.v = v

        self.contacts_active = {}
        for name in conf.contact_frames:
            self.contacts_active[name] = True

#        com_pos = np.empty((3, N_SIMULATION))*nan
#        com_vel = np.empty((3, N_SIMULATION))*nan
#        com_acc = np.empty((3, N_SIMULATION))*nan
#        com_pos_ref = np.empty((3, N_SIMULATION))*nan
#        com_vel_ref = np.empty((3, N_SIMULATION))*nan
#        com_acc_ref = np.empty((3, N_SIMULATION))*nan
#        # acc_des = acc_ref - Kp*pos_err - Kd*vel_err
#        com_acc_des = np.empty((3, N_SIMULATION))*nan
        self.offset = com_offset + self.robot.com(self.formulation.data())
        self.two_pi_f = np.copy(com_frequency)
        self.amp = np.copy(com_amplitude)
        self.two_pi_f_amp = self.two_pi_f * self.amp
        self.two_pi_f_squared_amp = self.two_pi_f * self.two_pi_f_amp
        self.sampleCom = self.trajCom.computeNext()
        self.reset(q, v, 0.0)