def test_baxter_ik(self):
env = ParamSetup.setup_env()
baxter = ParamSetup.setup_baxter()
can = ParamSetup.setup_green_can(geom=(0.02, 0.25))
baxter_body = OpenRAVEBody(env, baxter.name, baxter.geom)
can_body = OpenRAVEBody(env, can.name, can.geom)
baxter_body.set_transparency(0.5)
can_body.set_transparency(0.5)
manip = baxter_body.env_body.GetManipulator('right_arm')
robot = baxter_body.env_body
can = can_body.env_body
dof = robot.GetActiveDOFValues()
#Open the Gripper so there won't be collisions between gripper and can
dof[9], dof[-1] = 0.02, 0.02
robot.SetActiveDOFValues(dof)
iktype = IkParameterizationType.Transform6D
thetas = np.linspace(0, np.pi * 2, 10)
target_trans = OpenRAVEBody.transform_from_obj_pose([0.9, -0.23, 0.93],
[0, 0, 0])
can_body.env_body.SetTransform(target_trans)
target_trans[:3, :3] = target_trans[:3, :3].dot(
matrixFromAxisAngle([0, np.pi / 2, 0])[:3, :3])
can_trans = target_trans
body_trans = np.eye(4)
"""
To check whether baxter ik model works, uncomment the following
"""
# env.SetViewer('qtcoin')
for theta in thetas:
can_trans[:3, :3] = target_trans[:3, :3].dot(
matrixFromAxisAngle([theta, 0, 0])[:3, :3])
solution = manip.FindIKSolutions(
IkParameterization(can_trans, iktype),
IkFilterOptions.CheckEnvCollisions)
if len(solution) > 0:
print "Solution found with pose and rotation:"
print OpenRAVEBody.obj_pose_from_transform(can_trans)
else:
print "Solution not found with pose and rotation:"
print OpenRAVEBody.obj_pose_from_transform(can_trans)
for sols in solution:
dof[10:17] = sols
robot.SetActiveDOFValues(dof)
time.sleep(.2)
body_trans[:3, :3] = can_trans[:3, :3].dot(
matrixFromAxisAngle([0, -np.pi / 2, 0])[:3, :3])
can_body.env_body.SetTransform(body_trans)
self.assertTrue(
np.allclose([0.9, -0.23, 0.93],
manip.GetTransform()[:3, 3]))
def test_ik_arm_pose(self):
env = Environment()
attrs = {"name": ["pr2"], "pose": [(0, 0, 0)], "_type": ["Robot"], "geom": [], "backHeight": [0.2], "lGripper": [0.5], "rGripper": [0.5]}
attrs["lArmPose"] = [(0,0,0,0,0,0,0)]
attrs["rArmPose"] = [(0,0,0,0,0,0,0)]
attr_types = {"name": str, "pose": matrix.Vector3d, "_type": str, "geom": PR2, "backHeight": matrix.Value, "lArmPose": matrix.Vector7d, "rArmPose": matrix.Vector7d, "lGripper": matrix.Value, "rGripper": matrix.Value}
robot = parameter.Object(attrs, attr_types)
# Set the initial arm pose so that pose is not close to joint limit
robot.lArmPose = np.array([[np.pi/4, np.pi/8, np.pi/2, -np.pi/2, np.pi/8, -np.pi/8, np.pi/2]]).T
robot.rArmPose = np.array([[-np.pi/4, np.pi/8, -np.pi/2, -np.pi/2, -np.pi/8, -np.pi/8, np.pi/2]]).T
attrs = {"name": ["can"], "geom": (0.04, 0.25), "pose": ["undefined"], "rotation": [(0, 0, 0)], "_type": ["Can"]}
attr_types = {"name": str, "geom": BlueCan, "pose": matrix.Vector3d, "rotation": matrix.Vector3d, "_type": str}
can = parameter.Object(attrs, attr_types)
can.pose = np.array([[5.77887566e-01, -1.26743678e-01, 8.37601627e-01]]).T
can.rotation = np.array([[np.pi/4, np.pi/4, np.pi/4]]).T
# Create openRAVEBody for each parameter
can_body = OpenRAVEBody(env, can.name, can.geom)
robot_body = OpenRAVEBody(env, robot.name, robot.geom)
# Set the poses and dof values for each body
can_body.set_pose(can.pose, can.rotation)
robot_body.set_pose(robot.pose)
robot_body.set_dof(robot.backHeight, robot.lArmPose, robot.lGripper, robot.rArmPose, robot.rGripper)
# Solve the IK solution
ik_arm = robot_body.ik_arm_pose(can.pose, can.rotation)[0]
robot_body.set_dof(ik_arm[:1], robot.lArmPose, robot.lGripper, ik_arm[1:], robot.rGripper)
robot_trans = robot_body.env_body.GetLink("r_gripper_tool_frame").GetTransform()
robot_pos = OpenRAVEBody.obj_pose_from_transform(robot_trans)
# resulted robot eepose should be exactly the same as that can pose
self.assertTrue(np.allclose(can.pose.flatten(), robot_pos[:3], atol = 1e-4))
self.assertTrue(np.allclose(can.rotation.flatten(), robot_pos[3:], atol = 1e-4))
"""
def get_robot_info(self, robot_body):
# Provide functionality of Obtaining Robot information
tool_link = robot_body.env_body.GetLink("right_gripper")
manip_trans = tool_link.GetTransform()
# This manip_trans is off by 90 degree
pose = OpenRAVEBody.obj_pose_from_transform(manip_trans)
robot_trans = OpenRAVEBody.get_ik_transform(pose[:3], pose[3:])
arm_inds = list(range(10, 17))
return robot_trans, arm_inds
def get_col_free_armPose_ik(pred, negated, t, plan):
ee_pose = OpenRAVEBody.obj_pose_from_transform(
body.env_body.GetManipulator('right_arm').GetTransform())
pos, rot = ee_pose[:3], ee_pose[3:]
while arm_pose is None and iteration < const.MAX_ITERATION_STEP:
# for i in range(const.NUM_RESAMPLES):
pos_bias = np.random.random_sample(
(3, )) * const.BIAS_RADIUS * 2 - const.BIAS_RADIUS
rot_bias = np.random.random_sample(
(3, )) * const.ROT_BIAS * 2 - const.ROT_BIAS
# print pos_bias, rot_bias, iteration
print pos_bias, rot_bias
iteration += 1
arm_pose = get_ik_from_pose(pos + pos_bias, rot + rot_bias,
body.env_body, 'right_arm')
if arm_pose is not None:
print iteration
body.set_dof({'rArmPose': arm_pose})
def sample_ee_from_target(self, target):
"""
Sample all possible EE Pose that pr2 can grasp with
target: parameter of type Target
return: list of tuple in the format of (ee_pos, ee_rot)
"""
possible_ee_poses = []
targ_pos, targ_rot = target.value.flatten(), target.rotation.flatten()
ee_pos = targ_pos.copy()
target_trans = OpenRAVEBody.transform_from_obj_pose(targ_pos, targ_rot)
# rotate can's local z-axis by the amount of linear spacing between 0 to 2pi
angle_range = np.linspace(0, np.pi * 2, num=SAMPLE_SIZE)
for rot in angle_range:
target_trans = OpenRAVEBody.transform_from_obj_pose(targ_pos, targ_rot)
# rotate new ee_pose around can's rotation axis
rot_mat = matrixFromAxisAngle([0, 0, rot])
ee_trans = target_trans.dot(rot_mat)
ee_rot = OpenRAVEBody.obj_pose_from_transform(ee_trans)[3:]
possible_ee_poses.append((ee_pos, ee_rot))
return possible_ee_poses
def get_ee_from_target(targ_pos, targ_rot):
"""
Sample all possible EE Pose that pr2 can grasp with
target_pos: position of target we want to sample ee_pose form
target_rot: rotation of target we want to sample ee_pose form
return: list of ee_pose tuple in the format of (ee_pos, ee_rot) around target axis
"""
possible_ee_poses = []
ee_pos = targ_pos.copy()
target_trans = OpenRAVEBody.transform_from_obj_pose(targ_pos, targ_rot)
# rotate can's local z-axis by the amount of linear spacing between 0 to 2pi
angle_range = np.linspace(0, PI * 2, num=const.EE_ANGLE_SAMPLE_SIZE)
for rot in angle_range:
target_trans = OpenRAVEBody.transform_from_obj_pose(targ_pos, targ_rot)
# rotate new ee_pose around can's rotation axis
rot_mat = matrixFromAxisAngle([0, 0, rot])
ee_trans = target_trans.dot(rot_mat)
ee_rot = OpenRAVEBody.obj_pose_from_transform(ee_trans)[3:]
possible_ee_poses.append((ee_pos, ee_rot))
return possible_ee_poses
def resample_rrt_planner(pred, netgated, t, plan):
startp, endp = pred.startp, pred.endp
robot = pred.robot
body = pred._param_to_body[robot].env_body
manip_trans = body.GetManipulator("right_arm").GetTransform()
pose = OpenRAVEBody.obj_pose_from_transform(manip_trans)
manip_trans = OpenRAVEBody.get_ik_transform(pose[:3], pose[3:])
gripper_direction = manip_trans[:3, :3].dot(np.array([-1, 1, 0]))
lift_direction = manip_trans[:3, :3].dot(np.array([0, 0, -1]))
active_dof = body.GetManipulator("right_arm").GetArmIndices()
attr_inds = OrderedDict()
res = []
pred_test = [not pred.test(k, negated) for k in range(20)]
resample_ts = np.where(pred_test)[0]
start, end = resample_ts[0] - 1, resample_ts[-1] + 1
rave_body = pred._param_to_body[pred.robot]
dof_value_map = {
"lArmPose": pred.robot.lArmPose[:, start],
"lGripper": 0.02,
"rArmPose": pred.robot.rArmPose[:, start],
"rGripper": 0.02
}
rave_body.set_dof(dof_value_map)
rave_body.set_pose([0, 0, pred.robot.pose[:, start][0]])
body = pred._param_to_body[pred.robot].env_body
active_dof = body.GetManipulator('right_arm').GetArmIndices()
r_arm = pred.robot.rArmPose
traj = get_rrt_traj(plan.env, body, active_dof, r_arm[:, start],
r_arm[:, end])
result = process_traj(traj, end - start)
body.SetActiveDOFs(range(18))
for time in range(start + 1, end):
robot_attr_name_val_tuples = [('rArmPose', result[:,
time - start - 1])]
add_to_attr_inds_and_res(time, attr_inds, res, pred.robot,
robot_attr_name_val_tuples)
return np.array(res), attr_inds
def sample_ee_from_target(self, target):
"""
Sample all possible EE Pose that pr2 can grasp with
target: parameter of type Target
return: list of tuple in the format of (ee_pos, ee_rot)
"""
possible_ee_poses = []
targ_pos, targ_rot = target.value.flatten(), target.rotation.flatten()
ee_pos = targ_pos.copy()
target_trans = OpenRAVEBody.transform_from_obj_pose(targ_pos, targ_rot)
# rotate can's local z-axis by the amount of linear spacing between 0 to 2pi
angle_range = np.linspace(0, np.pi * 2, num=SAMPLE_SIZE)
for rot in angle_range:
target_trans = OpenRAVEBody.transform_from_obj_pose(
targ_pos, targ_rot)
# rotate new ee_pose around can's rotation axis
rot_mat = matrixFromAxisAngle([0, 0, rot])
ee_trans = target_trans.dot(rot_mat)
ee_rot = OpenRAVEBody.obj_pose_from_transform(ee_trans)[3:]
possible_ee_poses.append((ee_pos, ee_rot))
return possible_ee_poses
def resample_eereachable(pred, negated, t, plan):
attr_inds, res = OrderedDict(), []
robot, rave_body = pred.robot, pred._param_to_body[pred.robot]
target_pos, target_rot = pred.ee_pose.value.flatten(
), pred.ee_pose.rotation.flatten()
body = rave_body.env_body
rave_body.set_pose([0, 0, robot.pose[0, t]])
# Resample poses at grasping time
grasp_arm_pose = get_ik_from_pose(target_pos, target_rot, body,
'right_arm')
add_to_attr_inds_and_res(t, attr_inds, res, robot,
[('rArmPose', grasp_arm_pose.copy())])
plan.sampling_trace.append({
'type': robot.get_type(),
'data': {
'rArmPose': grasp_arm_pose
},
'timestep': t,
'pred': pred,
'action': "grasp"
})
# Setting poses for environments to extract transform infos
dof_value_map = {
"lArmPose": robot.lArmPose[:, t].reshape((7, )),
"lGripper": 0.02,
"rArmPose": grasp_arm_pose,
"rGripper": 0.02
}
rave_body.set_dof(dof_value_map)
# Prepare grasping direction and lifting direction
manip_trans = body.GetManipulator("right_arm").GetTransform()
pose = OpenRAVEBody.obj_pose_from_transform(manip_trans)
manip_trans = OpenRAVEBody.get_ik_transform(pose[:3], pose[3:])
gripper_direction = manip_trans[:3, :3].dot(np.array([-1, 0, 0]))
lift_direction = manip_trans[:3, :3].dot(np.array([0, 0, -1]))
# Resample grasping and retreating traj
for i in range(const.EEREACHABLE_STEPS):
approach_pos = target_pos - gripper_direction / np.linalg.norm(
gripper_direction) * const.APPROACH_DIST * (3 - i)
# rave_body.set_pose([0,0,robot.pose[0, t-3+i]])
approach_arm_pose = get_ik_from_pose(approach_pos, target_rot, body,
'right_arm')
# rave_body.set_dof({"rArmPose": approach_arm_pose})
add_to_attr_inds_and_res(t - 3 + i, attr_inds, res, robot,
[('rArmPose', approach_arm_pose)])
retreat_pos = target_pos + lift_direction / np.linalg.norm(
lift_direction) * const.RETREAT_DIST * (i + 1)
# rave_body.set_pose([0,0,robot.pose[0, t+1+i]])
retreat_arm_pose = get_ik_from_pose(retreat_pos, target_rot, body,
'right_arm')
add_to_attr_inds_and_res(t + 1 + i, attr_inds, res, robot,
[('rArmPose', retreat_arm_pose)])
robot._free_attrs['rArmPose'][:, t - const.EEREACHABLE_STEPS:t +
const.EEREACHABLE_STEPS + 1] = 0
robot._free_attrs['pose'][:, t - const.EEREACHABLE_STEPS:t +
const.EEREACHABLE_STEPS + 1] = 0
return np.array(res), attr_inds
def resample_eereachable_rrt(pred, negated, t, plan, inv=False):
# Preparing the variables
attr_inds, res = OrderedDict(), []
robot, rave_body = pred.robot, pred.robot.openrave_body
target_pos, target_rot = pred.ee_pose.value.flatten(
), pred.ee_pose.rotation.flatten()
body = rave_body.env_body
manip_name = "right_arm"
active_dof = body.GetManipulator(manip_name).GetArmIndices()
active_dof = np.hstack([[0], active_dof])
# Make sure baxter is well positioned in the env
rave_body.set_pose([0, 0, robot.pose[:, t]])
rave_body.set_dof({
'lArmPose': robot.lArmPose[:, t].flatten(),
'rArmPose': robot.rArmPose[:, t].flatten(),
"lGripper": np.array([0.02]),
"rGripper": np.array([0.02])
})
for param in plan.params.values():
if not param.is_symbol() and param != robot:
param.openrave_body.set_pose(param.pose[:, t].flatten(),
param.rotation[:, t].flatten())
# Resample poses at grasping time
grasp_arm_pose = get_ik_from_pose(target_pos, target_rot, body, manip_name)
# When Ik infeasible
if grasp_arm_pose is None:
return None, None
add_to_attr_inds_and_res(t, attr_inds, res, robot,
[('rArmPose', grasp_arm_pose.copy()),
('pose', robot.pose[:, t])])
# Store sampled pose
plan.sampling_trace.append({
'type': robot.get_type(),
'data': {
'rArmPose': grasp_arm_pose
},
'timestep': t,
'pred': pred,
'action': "grasp"
})
# Prepare grasping direction and lifting direction
manip_trans = body.GetManipulator("right_arm").GetTransform()
pose = OpenRAVEBody.obj_pose_from_transform(manip_trans)
manip_trans = OpenRAVEBody.get_ik_transform(pose[:3], pose[3:])
if inv:
# inverse resample_eereachable used in putdown action
approach_dir = manip_trans[:3, :3].dot(np.array([0, 0, -1]))
retreat_dir = manip_trans[:3, :3].dot(np.array([-1, 0, 0]))
approach_dir = approach_dir / np.linalg.norm(
approach_dir) * const.APPROACH_DIST
retreat_dir = -retreat_dir / np.linalg.norm(
retreat_dir) * const.RETREAT_DIST
else:
# Normal resample eereachable used in grasp action
approach_dir = manip_trans[:3, :3].dot(np.array([-1, 0, 0]))
retreat_dir = manip_trans[:3, :3].dot(np.array([0, 0, -1]))
approach_dir = -approach_dir / np.linalg.norm(
approach_dir) * const.APPROACH_DIST
retreat_dir = retreat_dir / np.linalg.norm(
retreat_dir) * const.RETREAT_DIST
resample_failure = False
# Resample entire approaching and retreating traj
for i in range(const.EEREACHABLE_STEPS):
approach_pos = target_pos + approach_dir * (3 - i)
approach_arm_pose = get_ik_from_pose(approach_pos, target_rot, body,
'right_arm')
retreat_pos = target_pos + retreat_dir * (i + 1)
retreat_arm_pose = get_ik_from_pose(retreat_pos, target_rot, body,
'right_arm')
if approach_arm_pose is None or retreat_arm_pose is None:
resample_failure = True
add_to_attr_inds_and_res(t - 3 + i, attr_inds, res, robot,
[('rArmPose', approach_arm_pose)])
add_to_attr_inds_and_res(t + 1 + i, attr_inds, res, robot,
[('rArmPose', retreat_arm_pose)])
# Ik infeasible
if resample_failure:
plan.sampling_trace[-1]['reward'] = -1
return None, None
# lock the variables
robot._free_attrs['rArmPose'][:, t - const.EEREACHABLE_STEPS:t +
const.EEREACHABLE_STEPS + 1] = 0
robot._free_attrs['pose'][:, t - const.EEREACHABLE_STEPS:t +
const.EEREACHABLE_STEPS + 1] = 0
# finding initial pose
init_timestep, ref_index = 0, 0
for i in range(len(plan.actions)):
act_range = plan.actions[i].active_timesteps
if act_range[0] <= t <= act_range[1]:
init_timestep = act_range[0]
ref_index = i
if pred.ee_resample is True and ref_index > 0:
init_timestep = plan.actions[ref_index - 1].active_timesteps[0]
init_dof = robot.rArmPose[:, init_timestep].flatten()
init_dof = np.hstack([robot.pose[:, init_timestep], init_dof])
end_dof = robot.rArmPose[:, t - const.EEREACHABLE_STEPS].flatten()
end_dof = np.hstack([robot.pose[:, t - const.EEREACHABLE_STEPS], end_dof])
timesteps = t - const.EEREACHABLE_STEPS - init_timestep + 2
raw_traj = get_rrt_traj(plan.env, body, active_dof, init_dof, end_dof)
# Restore dof0
dof = body.GetActiveDOFValues()
dof[0] = 0
body.SetActiveDOFValues(dof)
# trajectory is infeasible
if raw_traj == None:
plan.sampling_trace[-1]['reward'] = -1
return None, None
# initailize feasible trajectory
result_traj = process_traj(raw_traj, timesteps).T[1:-1]
ts = 1
for traj in result_traj:
add_to_attr_inds_and_res(init_timestep + ts, attr_inds, res, robot,
[('rArmPose', traj[1:]), ('pose', traj[:1])])
ts += 1
pred.ee_resample = True
can = plan.params['can0']
can.openrave_body.set_pose(can.pose[:, t], can.rotation[:, t])
rave_body.set_dof({'rArmPose': robot.rArmPose[:, t]})
return np.array(res), attr_inds
def test_ik_arm_pose(self):
env = Environment()
attrs = {
"name": ["pr2"],
"pose": [(0, 0, 0)],
"_type": ["Robot"],
"geom": [],
"backHeight": [0.2],
"lGripper": [0.5],
"rGripper": [0.5]
}
attrs["lArmPose"] = [(0, 0, 0, 0, 0, 0, 0)]
attrs["rArmPose"] = [(0, 0, 0, 0, 0, 0, 0)]
attr_types = {
"name": str,
"pose": matrix.Vector3d,
"_type": str,
"geom": PR2,
"backHeight": matrix.Value,
"lArmPose": matrix.Vector7d,
"rArmPose": matrix.Vector7d,
"lGripper": matrix.Value,
"rGripper": matrix.Value
}
robot = parameter.Object(attrs, attr_types)
# Set the initial arm pose so that pose is not close to joint limit
robot.lArmPose = np.array([[
np.pi / 4, np.pi / 8, np.pi / 2, -np.pi / 2, np.pi / 8, -np.pi / 8,
np.pi / 2
]]).T
robot.rArmPose = np.array([[
-np.pi / 4, np.pi / 8, -np.pi / 2, -np.pi / 2, -np.pi / 8,
-np.pi / 8, np.pi / 2
]]).T
attrs = {
"name": ["can"],
"geom": (0.04, 0.25),
"pose": ["undefined"],
"rotation": [(0, 0, 0)],
"_type": ["Can"]
}
attr_types = {
"name": str,
"geom": BlueCan,
"pose": matrix.Vector3d,
"rotation": matrix.Vector3d,
"_type": str
}
can = parameter.Object(attrs, attr_types)
can.pose = np.array([[5.77887566e-01, -1.26743678e-01,
8.37601627e-01]]).T
can.rotation = np.array([[np.pi / 4, np.pi / 4, np.pi / 4]]).T
# Create openRAVEBody for each parameter
can_body = OpenRAVEBody(env, can.name, can.geom)
robot_body = OpenRAVEBody(env, robot.name, robot.geom)
# Set the poses and dof values for each body
can_body.set_pose(can.pose, can.rotation)
robot_body.set_pose(robot.pose)
robot_body.set_dof(robot.backHeight, robot.lArmPose, robot.lGripper,
robot.rArmPose, robot.rGripper)
# Solve the IK solution
ik_arm = robot_body.ik_arm_pose(can.pose, can.rotation)[0]
robot_body.set_dof(ik_arm[:1], robot.lArmPose, robot.lGripper,
ik_arm[1:], robot.rGripper)
robot_trans = robot_body.env_body.GetLink(
"r_gripper_tool_frame").GetTransform()
robot_pos = OpenRAVEBody.obj_pose_from_transform(robot_trans)
# resulted robot eepose should be exactly the same as that can pose
self.assertTrue(
np.allclose(can.pose.flatten(), robot_pos[:3], atol=1e-4))
self.assertTrue(
np.allclose(can.rotation.flatten(), robot_pos[3:], atol=1e-4))
"""
