def duck_demo(viewer=True):
connect(use_gui=viewer)
# * in this case, the duck obj is in millimeter, we scale it into meter
duck_body = create_obj(DUCK_OBJ_PATH, scale=1e-3, color=(0, 0, 1, 1))
# * zoom in so we can see it, this is optional
camera_base_pt = (0, 0, 0)
camera_pt = np.array(camera_base_pt) + np.array([0.1, -0.1, 0.1])
set_camera_pose(tuple(camera_pt), camera_base_pt)
cprint('hello duck! <ctrl+left mouse> to pan', 'green')
wait_for_user()
def visualize_picknplace_planning_results(pkg_json_path, save_file_path, viewer=False):
# * create robot and pb environment
(robot_urdf, base_link_name, tool_root_link_name, ee_link_name, ik_joint_names, disabled_self_collision_link_names), \
(workspace_urdf, workspace_robot_disabled_link_names) = get_picknplace_robot_data()
picknplace_end_effector_urdf = get_picknplace_end_effector_urdf()
picknplace_tcp_def = get_picknplace_tcp_def()
# parse with connect
connect(use_gui=viewer)
with HideOutput():
_ = load_pybullet(robot_urdf, fixed_base=True)
full_trajs = parse_saved_trajectory(save_file_path)
disconnect()
# visualize plan
if viewer:
cart_time_step = None
tr_time_step = None
display_picknplace_trajectories(robot_urdf, ik_joint_names,
pkg_json_path, full_trajs, tool_root_link_name,
ee_urdf=picknplace_end_effector_urdf, workspace_urdf=workspace_urdf, animate=True,
cart_time_step=cart_time_step, tr_time_step=tr_time_step)
def sequenced_picknplace_plan(assembly_pkg_json_path, solve_method='sparse_ladder_graph', viewer=False, scale=1e-3,
sample_time=5, sparse_time_out=2, jt_res=0.1, pos_step_size=0.01, ori_step_size=np.pi/16,
viz_inspect=False, warning_pause=False, save_dir=None, **kwargs):
"""call pychoreo's planner to solve for picknplace Cartesian & transition trajectories.
Robot setup is specified in coop_assembly.choreo_interface.robot_setup
Parameters
----------
assembly_pkg_json_path : [type]
[description]
solve_method : str, optional
[description], by default 'sparse_ladder_graph'
viewer : bool, optional
[description], by default False
scale : [type], optional
[description], by default 1e-3
sample_time : int, optional
[description], by default 5
sparse_time_out : int, optional
[description], by default 2
jt_res : float, optional
joint resolution for transition planning, by default 0.1
pos_step_size : float, optional
interpolation step size for the end effector Cartesia movements, by default 0.01 (meter)
ori_step_size : [type], optional
[description], by default np.pi/16
viz_inspect : bool, optional
visualize planning process in a pybullet window (slow!), by default False
warning_pause : bool, optional
wait for user input if any of the planning process cannot find a solution, by default False
save_dir : string, optional
absolute directory path to save the planning result json file, by default None
Returns
-------
[type]
[description]
Raises
------
ValueError
[description]
"""
# TODO: assert solve method in avaiable list
# * load robot setup data
(robot_urdf, base_link_name, tool_root_link_name, ee_link_name, ik_joint_names, disabled_self_collision_link_names), \
(workspace_urdf, workspace_robot_disabled_link_names) = get_picknplace_robot_data()
picknplace_end_effector_urdf = get_picknplace_end_effector_urdf()
picknplace_tcp_def = get_picknplace_tcp_def()
# * create robot and pb environment
connect(use_gui=viewer)
# * adjust camera pose (optional)
if has_gui():
camera_base_pt = (0,0,0)
camera_pt = np.array(camera_base_pt) + np.array([1, 0, 0.5])
set_camera_pose(tuple(camera_pt), camera_base_pt)
with HideOutput():
# * pybullet can handle ROS-package path URDF automatically now (ver 2.5.7)!
robot = load_pybullet(robot_urdf, fixed_base=True)
workspace = load_pybullet(workspace_urdf, fixed_base=True)
# ee_body = create_obj(picknplace_end_effector_urdf)
ee_body = load_pybullet(picknplace_end_effector_urdf)
# * set robot idle configuration
ik_joints = joints_from_names(robot, ik_joint_names)
robot_start_conf = get_robot_init_conf()
set_joint_positions(robot, ik_joints, robot_start_conf)
# * create tool and tool TCP from flange (tool0) transformation
root_link = link_from_name(robot, tool_root_link_name)
# create end effector body
ee_attach = Attachment(robot, root_link, unit_pose(), ee_body)
# set up TCP transformation, just a renaming here
root_from_tcp = picknplace_tcp_def
if has_gui() :
# draw_tcp pose
ee_attach.assign()
ee_link_pose = get_pose(ee_attach.child)
draw_pose(multiply(ee_link_pose, root_from_tcp))
# * specify ik fn wrapper
ik_fn = IK_MODULE.get_ik
def get_sample_ik_fn(robot, ik_fn, ik_joint_names, base_link_name, tool_from_root=None):
def sample_ik_fn(world_from_tcp):
if tool_from_root:
world_from_tcp = multiply(world_from_tcp, tool_from_root)
return sample_tool_ik(ik_fn, robot, ik_joint_names, base_link_name, world_from_tcp, get_all=True) #,sampled=[0])
return sample_ik_fn
# ik generation function stays the same for all cartesian processes
sample_ik_fn = get_sample_ik_fn(robot, ik_fn, ik_joint_names, base_link_name, invert(root_from_tcp))
# * load shape & collision data
with open(assembly_pkg_json_path, 'r') as f:
json_data = json.loads(f.read())
assembly = Assembly.from_package(json_data)
elements = assembly.elements
for element in elements.values():
for unit_geo in element.unit_geometries:
unit_geo.rescale(scale)
# TODO: scale derived from the assembly package unit
# static_obstacles = []
# for ug in assembly.static_obstacle_geometries.values():
# static_obstacles.extend(ug.mesh)
# * load precomputed sequence / use assigned sequence
# TODO: load this as function argument
# element_seq = elements.keys()
element_seq = [0, 1, 2, 3, 4, 5]
# element_seq = [3, 4, 5]
print('sequence: ', element_seq)
# visualize goal pose
if has_gui():
# viz_len = 0.003
with WorldSaver():
for e_id in element_seq:
element = elements[e_id]
with LockRenderer():
print('e_id #{} : {}'.format(e_id, element))
for unit_geo in element.unit_geometries:
for pb_geo in unit_geo.pybullet_bodies:
set_pose(pb_geo, random.choice(unit_geo.get_goal_frames(get_pb_pose=True)))
# print('---------')
# wait_for_user()
# wait_for_user()
# * construct ignored body-body links for collision checking
# in this case, including self-collision between links of the robot
disabled_self_collisions = get_disabled_collisions(robot, disabled_self_collision_link_names)
# and links between the robot and the workspace (e.g. robot_base_link to base_plate)
extra_disabled_collisions = get_body_body_disabled_collisions(robot, workspace, workspace_robot_disabled_link_names)
# TODO: extra disabled collisions as function argument
extra_disabled_collisions.update({
((robot, link_from_name(robot, 'robot_link_5')), (ee_body, link_from_name(ee_body, 'eef_base_link'))),
((robot, link_from_name(robot, 'robot_link_6')), (ee_body, link_from_name(ee_body, 'eef_base_link')))
})
# * create cartesian processes without a sequence being given, with random pose generators
cart_process_seq = build_picknplace_cartesian_process_seq(
element_seq, elements,
robot, ik_joint_names, root_link, sample_ik_fn,
ee_attachs=[ee_attach], self_collisions=True, disabled_collisions=disabled_self_collisions,
obstacles=[workspace],extra_disabled_collisions=extra_disabled_collisions,
tool_from_root=invert(root_from_tcp), viz_step=False, pick_from_same_rack=True,
pos_step_size=pos_step_size, ori_step_size=ori_step_size)
# specifically for UR5, because of its wide joint range, we need to apply joint value snapping
for cp in cart_process_seq:
cp.target_conf = robot_start_conf
with LockRenderer(not viz_inspect):
if solve_method == 'ladder_graph':
print('\n'+'#' * 10)
print('Solving with the vanilla ladder graph search algorithm.')
cart_process_seq = solve_ladder_graph_from_cartesian_process_list(cart_process_seq,
verbose=True, warning_pause=warning_pause, viz_inspect=viz_inspect, check_collision=False, start_conf=robot_start_conf)
elif solve_method == 'sparse_ladder_graph':
print('\n'+'#' * 10)
print('Solving with the sparse ladder graph search algorithm.')
sparse_graph = SparseLadderGraph(cart_process_seq)
sparse_graph.find_sparse_path(verbose=True, vert_timeout=sample_time, sparse_sample_timeout=sparse_time_out)
cart_process_seq = sparse_graph.extract_solution(verbose=True, start_conf=robot_start_conf)
else:
raise ValueError('Invalid solve method!')
assert all(isinstance(cp, CartesianProcess) for cp in cart_process_seq)
pnp_trajs = [[] for _ in range(len(cart_process_seq))]
for cp_id, cp in enumerate(cart_process_seq):
element_attachs = []
for sp_id, sp in enumerate(cp.sub_process_list):
assert sp.trajectory, '{}-{} does not have a Cartesian plan found!'.format(cp, sp)
# ! reverse engineer the grasp pose
if sp.trajectory.tag == 'pick_retreat':
unit_geo = elements[sp.trajectory.element_id].unit_geometries[0]
e_bodies = unit_geo.pybullet_bodies
for e_body in e_bodies:
set_pose(e_body, unit_geo.get_initial_frames(get_pb_pose=True)[0])
set_joint_positions(sp.trajectory.robot, sp.trajectory.joints, sp.trajectory.traj_path[0])
element_attachs.append(create_attachment(sp.trajectory.robot, root_link, e_body))
if sp.trajectory.tag == 'pick_retreat' or sp.trajectory.tag == 'place_approach':
sp.trajectory.attachments= element_attachs
pnp_trajs[cp_id].append(sp.trajectory)
full_trajs = pnp_trajs
# * transition motion planning between extrusions
return2idle = True
transition_traj = solve_transition_between_picknplace_processes(pnp_trajs, elements, robot_start_conf,
disabled_collisions=disabled_self_collisions,
extra_disabled_collisions=extra_disabled_collisions,
obstacles=[workspace], return2idle=return2idle,
resolutions=[jt_res]*len(ik_joints),
**kwargs)
# * weave the Cartesian and transition processses together
for cp_id, print_trajs in enumerate(full_trajs):
print_trajs.insert(0, transition_traj[cp_id][0])
print_trajs.insert(3, transition_traj[cp_id][1])
if return2idle:
full_trajs[-1].append(transition_traj[-1][-1])
if save_dir is None:
here = os.path.dirname(__file__)
save_dir = os.path.join(here, 'results')
export_trajectory(save_dir, full_trajs, ee_link_name, indent=1, shape_file_path=assembly_pkg_json_path,
include_robot_data=True, include_link_path=True)
# * disconnect and close pybullet engine used for planning, visualizing trajectories will start a new one
reset_simulation()
disconnect()
if viewer:
cart_time_step = None
tr_time_step = None
display_picknplace_trajectories(robot_urdf, ik_joint_names,
assembly_pkg_json_path, full_trajs, tool_root_link_name,
ee_urdf=picknplace_end_effector_urdf, workspace_urdf=workspace_urdf, animate=True,
cart_time_step=cart_time_step, tr_time_step=tr_time_step)
def descartes_demo(viewer=True, scaling_test=False, draw_graph=True):
connect(use_gui=viewer)
with HideOutput():
robot = load_pybullet(KUKA_ROBOT_URDF, fixed_base=True)
# * adjust camera pose (optional)
# has_gui checks if the GUI mode is enabled
if has_gui():
camera_base_pt = (0,0,0)
camera_pt = np.array(camera_base_pt) + np.array([1, -0.5, 1])
set_camera_pose(tuple(camera_pt), camera_base_pt)
cprint('Hello robot world! <ctrl+left mouse> to pan', 'green')
wait_if_gui()
ik_joints = get_movable_joints(robot)
ik_joint_names = get_joint_names(robot, ik_joints)
print('ik joint names: {}'.format(ik_joint_names))
lower_limits, upper_limits = get_custom_limits(robot, ik_joints)
print('joint lower limit: {}'.format(lower_limits))
print('joint upper limit: {}'.format(upper_limits))
# we can also read these velocities from the SRDF file (using e.g. COMPAS_FAB)
# I'm a bit lazy, just handcode the values here
vel_limits = {0 : 6.28318530718,
1 : 5.23598775598,
2 : 6.28318530718,
3 : 6.6497044501,
4 : 6.77187749774,
5 : 10.7337748998}
# set the robot to a "comfortable" start configuration, optional
robot_start_conf = [0,-np.pi/2,np.pi/2,0,0,0]
set_joint_positions(robot, ik_joints, robot_start_conf)
tool_link = link_from_name(robot, EE_LINK_NAME)
robot_base_link = link_from_name(robot, ROBOT_BASE_LINK_NAME)
# tool_from_root = get_relative_pose(robot, root_link, tool_link)
# * draw EE pose
if has_gui() :
tcp_pose = get_link_pose(robot, tool_link)
draw_pose(tcp_pose)
circle_center = np.array([0.6, 0, 0.2])
circle_r = 0.1
# * generate multiple circles
ee_poses = []
full_angle = 2*2*np.pi
# total num of path pts, one path point per 5 degree
n_pt = int(full_angle / (np.pi/180 * 5))
# full_angle = np.pi
for a in np.linspace(0.0, full_angle, num=n_pt):
pt = circle_center + circle_r*np.array([np.cos(a), np.sin(a), 0])
circ_pose = multiply(Pose(point=pt, euler=Euler(yaw=a+np.pi/2)), Pose(euler=Euler(roll=np.pi*3/4)))
draw_pose(circ_pose, length=0.01)
ee_poses.append(circ_pose)
# def get_ee_sample_fn(roll_gen, pitch_gen, yaw_gen):
# def ee_sample_fn(ee_pose):
# # a finite generator
# for roll, pitch, yaw in product(roll_gen, pitch_gen, yaw_gen):
# yield multiply(ee_pose, Pose(euler=Euler(roll=roll, pitch=pitch, yaw=yaw)))
# return ee_sample_fn
# roll_sample_size = 3
# pitch_sample_size = 3
# yaw_sample_size = 1
# delta_roll = np.pi/12
# delta_pitch = np.pi/12
# roll_gen = np.linspace(-delta_roll, delta_roll, num=roll_sample_size)
# pitch_gen = np.linspace(-delta_pitch, delta_pitch, num=pitch_sample_size)
# yaw_gen = np.linspace(0.0, 2*np.pi, num=yaw_sample_size)
# ee_sample_fn = get_ee_sample_fn(roll_gen, pitch_gen, yaw_gen)
# for ep in ee_sample_fn(ee_poses[0]):
# draw_pose(ep, length=0.03)
# * baseline, keeping the EE z axis rotational dof fixed
st_time = time.time()
path = plan_cartesian_motion(robot, robot_base_link, tool_link, ee_poses)
print('Solving time: {}'.format(elapsed_time(st_time)))
if path is None:
cprint('Gradient-based ik cartesian planning cannot find a plan!', 'red')
else:
cprint('Gradient-based ik cartesian planning find a plan!', 'green')
time_step = 0.03
for conf in path:
set_joint_positions(robot, ik_joints, conf)
wait_for_duration(time_step)
if draw_graph:
print('drawing graph...')
plot_joint_val(path, lower_limits, upper_limits, method='GradIK')
print('='*20)
# * Now, let's try if using ladder graph without releasing the ee dof can give us a good trajectory
# you should be doing something similar in your picture, right?
# First, we will need ikfast to obtain ik solution variance, same in Descartes
ik_fn = ikfast_kuka_kr6_r900.get_ik
# we have to specify ik fn wrapper and feed it into pychoreo
def get_sample_ik_fn(robot, ik_fn, robot_base_link, ik_joints, tool_from_root=None):
def sample_ik_fn(world_from_tcp):
if tool_from_root:
world_from_tcp = multiply(world_from_tcp, tool_from_root)
return sample_tool_ik(ik_fn, robot, ik_joints, world_from_tcp, robot_base_link, get_all=True)
return sample_ik_fn
# ik generation function stays the same for all cartesian processes
sample_ik_fn = get_sample_ik_fn(robot, ik_fn, robot_base_link, ik_joints)
# we ignore self collision in this tutorial, the collision_fn only considers joint limit now
# See : https://github.com/yijiangh/pybullet_planning/blob/dev/tests/test_collisions.py
# for more info on examples on using collision function
collision_fn = get_collision_fn(robot, ik_joints, obstacles=[],
attachments=[], self_collisions=False,
# disabled_collisions=disabled_collisions,
# extra_disabled_collisions=extra_disabled_collisions,
custom_limits={})
# Let's check if our ik sampler is working properly
# uncomment the following if you want to play with this
# for p in ee_poses:
# print('-'*5)
# pb_q = inverse_kinematics(robot, tool_link, p)
# if pb_q is None:
# cprint('pb ik can\'t find an ik solution', 'red')
# qs = sample_ik_fn(p)
# if qs is not None:
# cprint('But Ikfast does find one! {}'.format(qs[0]), 'green')
# # set_joint_positions(robot, ik_joints, qs[0])
# # wait_if_gui()
# else:
# cprint('ikfast can\'t find an ik solution', 'red')
# * Ok, now we have the ik sampler and collision function ready, let's see if we can find a valid cartesian trajectory!
ee_vel = 0.005 # m/s
# st_time = time.time()
# path, cost = plan_cartesian_motion_lg(robot, ik_joints, ee_poses, sample_ik_fn, collision_fn, \
# custom_vel_limits=vel_limits, ee_vel=ee_vel)
# print('Solving time: {}'.format(elapsed_time(st_time)))
# if path is None:
# cprint('ladder graph (w/o releasing dof) cartesian planning cannot find a plan!', 'red')
# else:
# cprint('ladder graph (w/o releasing dof) cartesian planning find a plan!', 'green')
# cprint('Cost: {}'.format(cost), 'yellow')
# time_step = 0.03
# for conf in path:
# # cprint('conf: {}'.format(conf))
# set_joint_positions(robot, ik_joints, conf)
# wait_for_duration(time_step)
# if draw_graph:
# print('drawing graph...')
# plot_joint_val(path, lower_limits, upper_limits, method='IKFast+LadderGraph')
# print('='*20)
# * Now, let's see if releasing EE z axis dof can bring down the cost
# First, let's build an end effector pose sampler to release the EE z axis dof!
## * EE z axis dof generator
def get_ee_sample_fn(roll_gen, pitch_gen, yaw_gen):
def ee_sample_fn(ee_pose):
# a finite generator
for roll, pitch, yaw in product(roll_gen, pitch_gen, yaw_gen):
yield multiply(ee_pose, Pose(euler=Euler(roll=roll, pitch=pitch, yaw=yaw)))
return ee_sample_fn
# by increasing this number we can see the cost go down
roll_sample_size = 10
pitch_sample_size = 10
yaw_sample_size = 10
delta_roll = np.pi/6
delta_pitch = np.pi/6
delta_yaw = np.pi/6
roll_gen = np.linspace(-delta_roll, delta_roll, num=roll_sample_size)
pitch_gen = np.linspace(-delta_pitch, delta_pitch, num=pitch_sample_size)
yaw_gen = np.linspace(-delta_yaw, delta_yaw, num=yaw_sample_size)
ee_sample_fn = get_ee_sample_fn(roll_gen, pitch_gen, yaw_gen)
st_time = time.time()
path, cost = plan_cartesian_motion_lg(robot, ik_joints, ee_poses, sample_ik_fn, collision_fn, sample_ee_fn=ee_sample_fn, \
custom_vel_limits=vel_limits, ee_vel=ee_vel)
print('Solving time: {}'.format(elapsed_time(st_time)))
# TODO: plot ee z axis deviation plot
# the ladder graph cost is just summation of all adjacent joint difference
# so the following assertion should be true
# conf_array = np.array(path)
# conf_diff = np.abs(conf_array[:-1,:] - conf_array[1:,:])
# np_cost = np.sum(conf_diff)
# assert np.allclose(np_cost, cost), '{} - {}'.format(np_cost, cost)
if path is None:
cprint('ladder graph (releasing EE z dof) cartesian planning cannot find a plan!', 'red')
else:
cprint('ladder graph (releasing EE z dof) cartesian planning find a plan!', 'cyan')
cprint('Cost: {}'.format(cost), 'yellow')
time_step = 0.03
for conf in path:
# cprint('conf: {}'.format(conf))
set_joint_positions(robot, ik_joints, conf)
wait_for_duration(time_step)
if draw_graph:
print('drawing graph...')
plot_joint_val(path, lower_limits, upper_limits, method='IKFast+LadderGraph+release z dof-res{}'.format(yaw_sample_size))
print('='*20)
# Now, let's do a scaling test:
# if scaling_test:
# res_sc = list(range(4,30))
# cost_sc = []
# for res in res_sc:
# yaw_sample_size = res
# yaw_gen = np.linspace(0.0, 2*np.pi, num=yaw_sample_size)
# ee_sample_fn = get_ee_sample_fn(yaw_gen)
# path, cost = plan_cartesian_motion_lg(robot, ik_joints, ee_poses, sample_ik_fn, collision_fn, sample_ee_fn=ee_sample_fn)
# assert path is not None
# cost_sc.append(cost)
# if res % 5 == 0:
# if draw_graph:
# plot_joint_val(path, lower_limits, upper_limits, method='IKFast+LadderGraph+release z dof-res{}'.format(yaw_sample_size))
# fig, ax = plt.subplots()
# ax.plot(res_sc, cost_sc)
# ax.set(xlabel='yaw angle discr resolution', ylabel='total joint difference cost',
# title='Total joint diff scaling test')
# ax.grid()
# fig.savefig(os.path.join(HERE, 'images', 'tota_diff-scaling.png'))
# # plt.show()
wait_if_gui('Press enter to exit')
def assembly_demo(viewer=True, debug=False):
connect(use_gui=viewer)
# * zoom in so we can see it, this is optional
camera_base_pt = (0, 0, 0)
camera_pt = np.array(camera_base_pt) + np.array([0.1, -0.1, 0.1])
set_camera_pose(tuple(camera_pt), camera_base_pt)
# * a simple example showing two artificial beam colliding at the interface
# compare this "brep-like" approach to the mesh approach below,
# here we have less numerical error at the interface
cprint('=' * 10)
cprint('Checking collisions between two created boxes.', 'cyan')
w = 0.05
l = 0.5
h = 0.01
eps_list = [-1e-14, 0, 1e-3, 1e-6, 1e-14]
b0_body = create_box(w, l, h, color=apply_alpha(RED, 0.5))
b1_body = create_box(w, l, h, color=apply_alpha(BLUE, 0.5))
for eps in eps_list:
print('=' * 5)
cprint('Pertubation distance: {}'.format(eps), 'yellow')
# the beam on the right is perturbed towards left by eps
set_pose(b0_body, Pose(point=[0, l / 2 - eps, 0]))
set_pose(b1_body, Pose(point=[0, -l / 2, 0]))
is_collided = pairwise_collision(b0_body, b1_body)
if is_collided:
cprint(
'Collision detected! The penetration depth shown below should be close to eps={}'
.format(eps), 'red')
cr = pairwise_collision_info(b0_body, b1_body)
draw_collision_diagnosis(cr, focus_camera=True)
else:
cprint('No collision detected between b0 and b1.', 'green')
assert eps <= 0 or is_collided
# * now let's load Kathrin's beams and check pairwise collision among them
# notice that in pybullet everything is in METER, and each beam is configured at its pose in the world
# already in the obj files (directly exported from Rhino).
cprint('=' * 10)
cprint('Checking collisions among loaded obj elements.', 'cyan')
beam_from_names = {}
for i in range(150):
beam_path = os.path.join(ASSEMBLY_OBJ_DIR, 'element_{}.obj'.format(i))
beam_from_names[i] = create_obj(beam_path,
scale=1,
color=apply_alpha(BLUE, 0.2))
collided_pairs = set()
for i in range(150):
for j in range(i + 1, 150):
if (i, j) not in collided_pairs and (j, i) not in collided_pairs:
if pairwise_collision(beam_from_names[i], beam_from_names[j]):
# using meshes (comparing to the `create_box` approach above) induces numerical errors
# in our case here the touching faces will be checked as collisions.
# Thus, we have to query the getClosestPoint info and use the penetration depth to filter these "touching" cases out
# reference for these info: https://docs.google.com/document/d/10sXEhzFRSnvFcl3XxNGhnD4N2SedqwdAvK3dsihxVUA/edit#heading=h.cb0co8y2vuvc
cr = body_collision_info(beam_from_names[i],
beam_from_names[j])
penetration_depth = get_distance(cr[0][5], cr[0][6])
# this numner `3e-3` below is based on some manual experiement,
# might need to be changed accordingly for specific scales and input models
if penetration_depth > 3e-3:
cprint(
'({}-{}) colliding : penetrating depth {:.4E}'.
format(i, j, penetration_depth), 'red')
collided_pairs.add((i, j))
if debug:
draw_collision_diagnosis(cr, focus_camera=False)
# all the colliding information is in `collided_pairs`
cprint(
'In total we\'ve found {} collided element pairs!'.format(
len(collided_pairs)), 'green')
wait_if_gui('End.')
def ur_demo(viewer=True, robot_path=UR_ROBOT_URDF, ee_path=EE_PATH, \
workspace_path=MIT_WORKSPACE_PATH, attach_obj_path=ATTACH_OBJ_PATH, obstacle_obj_path=OBSTACLE_OBJ_PATH):
# * this will create the pybullet GUI
# setting viewers=False will enter GUI-free mode
connect(use_gui=viewer)
cprint(
"Welcome to pybullet! <Ctrl+left mouse> to rotate, <Ctrl+middle mouse> to move the camera, <Ctrl+right mouse> to zoom",
'green')
cprint('But the space is empty, let\'s load our robot!', 'yellow')
# wait_for_user is your friend! It will pause the console, but having a separate thread keeping
# the GUI running so you can rotate and see
wait_for_user()
# * This is how we load a robot from a URDF, a workspace from a URDF, or simply a mesh object from an obj file
# Notice that the pybullet uses *METER* by default, make sure you scale things properly!
robot = load_pybullet(robot_path, fixed_base=True)
workspace = load_pybullet(workspace_path, fixed_base=True)
ee_body = create_obj(ee_path)
# this will print all the bodies' information in your console
dump_world()
cprint(
'You just loaded a robot, a workspace (with many static objects as its link, I modeled our good old MIT 3-412 shop here), '
+ 'and an end effector (it\'s inside the robot base now)', 'green')
wait_for_user()
# * adjust camera pose (optional)
# has_gui checks if the GUI mode is enabled
if has_gui():
camera_base_pt = (0, 0, 0)
camera_pt = np.array(camera_base_pt) + np.array([1, -0.5, 0.5])
set_camera_pose(tuple(camera_pt), camera_base_pt)
# * each joint of the robot are assigned with an integer in pybullet
ik_joints = get_movable_joints(robot)
ik_joint_names = get_joint_names(robot, ik_joints)
cprint('Joint {} \ncorresponds to:\n{}'.format(ik_joints, ik_joint_names),
'green')
robot_start_conf = [0, -1.65715, 1.71108, -1.62348, 0, 0]
cprint("This is before updating pose", 'yellow')
wait_for_user()
# * set joint configuration, the robot's pose will be updated
set_joint_positions(robot, ik_joints, robot_start_conf)
cprint("This is after set joint pose: {}".format(robot_start_conf),
'green')
wait_for_user()
tool_attach_link_name = 'ee_link'
tool_attach_link = link_from_name(robot, tool_attach_link_name)
# * attach the end effector
ee_link_pose = get_link_pose(robot, tool_attach_link)
set_pose(ee_body, ee_link_pose)
ee_attach = create_attachment(robot, tool_attach_link, ee_body)
# we need to call "assign()" to update the attachment to the current end effector pose
ee_attach.assign()
# let's load a bar element (obj) and a box (pybullet primitive shape) into the world
attached_bar_body = create_obj(attach_obj_path)
box_body = create_obj(obstacle_obj_path)
cprint('We loaded a box to our scene!', 'green')
wait_for_user()
# * attach the bar
ee_link_from_tcp = Pose(point=(0.094, 0, 0))
set_pose(attached_bar_body, multiply(ee_link_pose, ee_link_from_tcp))
bar_attach = create_attachment(robot, tool_attach_link, attached_bar_body)
bar_attach.assign()
cprint('The bar element is attached to the robot', 'green')
wait_for_user()
attachments = [ee_attach, bar_attach]
# * Let's do some collision checking
# * specify disabled link pairs for collision checking (because they are adjacent / impossible to collide)
# link name corresponds to the ones specified in the URDF
# again, each robot link is assigned with an integer index in pybullet
robot_self_collision_disabled_link_names = [
('base_link', 'shoulder_link'), ('ee_link', 'wrist_1_link'),
('ee_link', 'wrist_2_link'), ('ee_link', 'wrist_3_link'),
('forearm_link', 'upper_arm_link'), ('forearm_link', 'wrist_1_link'),
('shoulder_link', 'upper_arm_link'), ('wrist_1_link', 'wrist_2_link'),
('wrist_1_link', 'wrist_3_link'), ('wrist_2_link', 'wrist_3_link')
]
self_collision_links = get_disabled_collisions(
robot, robot_self_collision_disabled_link_names)
cprint('self_collision_links disabled: {}'.format(self_collision_links),
'yellow')
extra_disabled_link_names = [('base_link', 'MIT_3412_robot_base_plate'),
('shoulder_link', 'MIT_3412_robot_base_plate')
]
extra_disabled_collisions = get_body_body_disabled_collisions(
robot, workspace, extra_disabled_link_names)
cprint('extra disabled: {}'.format(extra_disabled_collisions), 'yellow')
print('#' * 10)
cprint('Checking robot links self-collision', 'green')
collision_fn = get_collision_fn(robot,
ik_joints,
obstacles=[],
attachments=attachments,
self_collisions=True,
disabled_collisions=self_collision_links)
conf = [-1.029744, -1.623156, 2.844887, -0.977384, 1.58825, 0.314159]
# self collision, this should return true
# this function will first set the robot's joint configuration (together with the attachment)
# and check collision
# notice that turning on the diagnosis flag here will show you where the collision is happening
cprint(
'Notice that the diagnosis mode will zoom the camera to where the collision is detected. Move the camera around if you can\'t see what\'s going on there.',
'yellow')
assert collision_fn(conf, diagnosis=True)
print('#' * 10)
cprint('Checking robot links - holding attachment self-collision', 'green')
collision_fn = get_collision_fn(
robot,
ik_joints,
obstacles=[],
attachments=attachments,
self_collisions=True,
disabled_collisions=self_collision_links,
extra_disabled_collisions=extra_disabled_collisions)
conf = [0.03500, -2.26900, 2.44300, 1.117, 1.6579, 0.105]
assert collision_fn(conf, diagnosis=True)
print('\n')
print('#' * 10)
cprint('Checking robot links to obstacles (w/o links) collision', 'green')
collision_fn = get_collision_fn(
robot,
ik_joints,
obstacles=[box_body],
attachments=attachments,
self_collisions=True,
disabled_collisions=self_collision_links,
extra_disabled_collisions=extra_disabled_collisions)
conf = [
-0.105, -0.76800000000000002, 1.292, -0.61099999999999999, 1.484, 0.105
]
assert collision_fn(conf, diagnosis=True)
print('\n')
print('#' * 10)
cprint('Checking robot links to multi-link obstacle collision', 'green')
collision_fn = get_collision_fn(
robot,
ik_joints,
obstacles=[workspace],
attachments=[],
self_collisions=True,
disabled_collisions=self_collision_links,
extra_disabled_collisions=extra_disabled_collisions)
conf = [
-0.17499999999999999, -3.194, 0.33200000000000002, -1.6579999999999999,
1.431, 0.105
]
assert collision_fn(conf, diagnosis=True)
print('\n')
print('#' * 10)
cprint('Checking attachment to obstacles (w/o links) collision', 'green')
collision_fn = get_collision_fn(
robot,
ik_joints,
obstacles=[workspace, box_body],
attachments=attachments,
self_collisions=True,
disabled_collisions=self_collision_links,
extra_disabled_collisions=extra_disabled_collisions)
conf = [
-2.8100000000000001, -1.484, -1.9199999999999999, -1.6579999999999999,
1.431, 0.105
]
assert collision_fn(conf, diagnosis=True)
print('\n')
print('#' * 10)
cprint('Checking attachment to multi-link obstacle collision', 'green')
collision_fn = get_collision_fn(
robot,
ik_joints,
obstacles=[workspace],
attachments=attachments,
self_collisions=True,
disabled_collisions=self_collision_links,
extra_disabled_collisions=extra_disabled_collisions)
conf = [
-0.17499999999999999, -2.4780000000000002, 0.33200000000000002,
-1.6579999999999999, 1.431, 0.105
]
assert collision_fn(conf, diagnosis=True)
print('\n')
# * collision checking exoneration
print('#' * 10)
print('self-link collision disable')
collision_fn = get_collision_fn(robot,
ik_joints,
obstacles=[],
attachments=[],
self_collisions=False)
conf = [-1.029744, -1.623156, 2.844887, -0.977384, 1.58825, 0.314159]
assert not collision_fn(conf, diagnosis=True)
print('\n')
print('#' * 10)
cprint('robot links to obstacle collision exoneration', 'green')
cprint(
'In this example, the first collision function will check collision between the robot and the box, '
+ 'but the second collision function will ignore it.', 'yellow')
collision_fn = get_collision_fn(
robot,
ik_joints,
obstacles=[box_body],
attachments=[],
self_collisions=True,
disabled_collisions=self_collision_links,
)
collision_fn_disable = get_collision_fn(
robot,
ik_joints,
obstacles=[box_body],
attachments=[],
self_collisions=True,
disabled_collisions=self_collision_links,
extra_disabled_collisions=extra_disabled_collisions.union([
((robot, link_from_name(robot,
'forearm_link')), (box_body, BASE_LINK))
]),
)
conf = [
-3.2639999999999998, -2.6880000000000002, -0.85499999999999998, -1.536,
3.0369999999999999, -0.070000000000000007
]
assert collision_fn(conf, diagnosis=True)
assert not collision_fn_disable(conf, diagnosis=True)
print('\n')
# * joint value overflow checking & exoneration
cprint('joint value overflow checking & exoneration', 'green')
cprint(
'collision_fn also checks for robot\'s joint limit as well. We can also exonerate it by passing custom_limits into the collision_fn',
'yellow')
def get_custom_limits_from_name(robot, joint_limits):
return {
joint_from_name(robot, joint): limits
for joint, limits in joint_limits.items()
}
custom_limits = get_custom_limits_from_name(robot, {
'shoulder_pan_joint': (-7.9, 0),
'elbow_joint': (-8.0, 0)
})
collision_fn = get_collision_fn(robot, ik_joints)
collision_fn_disable = get_collision_fn(robot,
ik_joints,
custom_limits=custom_limits)
conf = [
-7.8450000000000002, -2.1469999999999998, -7.99, -0.92500000000000004,
1.78, 0.105
]
assert collision_fn(conf, diagnosis=True)
assert not collision_fn_disable(conf, diagnosis=True)
print('\n')
def rfl_demo(viewer=True):
arm = 'right'
connect(use_gui=viewer)
robot = load_pybullet(RFL_ROBOT_URDF, fixed_base=True)
set_camera(yaw=-90, pitch=-40, distance=10, target_position=(0, 7.5, 0))
cprint('hello RFL! <ctrl+left mouse> to pan', 'green')
wait_for_user()
# create a box to be picked up
# see: https://pybullet-planning.readthedocs.io/en/latest/reference/generated/pybullet_planning.interfaces.env_manager.create_box.html#pybullet_planning.interfaces.env_manager.create_box
block = create_box(0.059, 20., 0.089)
block_x = 10
block_y = 5.
block_z = 1.5
set_pose(
block,
Pose(Point(x=block_x, y=block_y, z=block_z), Euler(yaw=np.pi / 2)))
arm_tag = arm[0]
arm_joint_names = ['gantry_x_joint',
'{}_gantry_y_joint'.format(arm_tag),
'{}_gantry_z_joint'.format(arm_tag)] + \
['{}_robot_joint_{}'.format(arm_tag, i+1) for i in range(6)]
arm_joints = joints_from_names(robot, arm_joint_names)
# * if a subset of joints is used, use:
# arm_joints = joints_from_names(robot, arm_joint_names[1:]) # this will disable the gantry-x joint
cprint('Used joints: {}'.format(get_joint_names(robot, arm_joints)),
'yellow')
# * get a joint configuration sample function:
# it separately sample each joint value within the feasible range
sample_fn = get_sample_fn(robot, arm_joints)
cprint(
'Randomly sample robot configuration and set them! (no collision check is performed)',
'blue')
wait_for_user()
for i in range(5):
# randomly sample a joint conf
sampled_conf = sample_fn()
set_joint_positions(robot, arm_joints, sampled_conf)
cprint('#{} | Conf sampeld: {}'.format(i, sampled_conf), 'green')
wait_for_user()
# * now let's plan a trajectory
# we use y-z-arm 6 joint all together here
cprint(
'Randomly sample robot start/end configuration and comptue a motion plan! (no self-collision check is performed)',
'blue')
print(
'Disabled collision links needs to be given (can be parsed from a SRDF via compas_fab)'
)
for _ in range(5):
print('=' * 10)
q1 = list(sample_fn())
# intentionly make the robot needs to cross the collision object
# let it start from the right side
q1[0] = 0.
q1[1] = 0
set_joint_positions(robot, arm_joints, q1)
cprint('Sampled start conf: {}'.format(q1), 'cyan')
wait_for_user()
# let it ends at the left side
q2 = list(sample_fn())
q2[0] = 0.5
q2[1] = 7.0
cprint('Sampled end conf: {}'.format(q2), 'cyan')
path = plan_joint_motion(robot,
arm_joints,
q2,
obstacles=[block],
self_collisions=False,
custom_limits={arm_joints[0]: [0.0, 1.2]})
if path is None:
cprint('no plan found', 'red')
continue
else:
wait_for_user(
'a motion plan is found! Press enter to start simulating!')
# adjusting this number will adjust the simulation speed
time_step = 0.03
for conf in path:
set_joint_positions(robot, arm_joints, conf)
wait_for_duration(time_step)
def test_import():
connect(use_gui=True)
def panda_demo(viewer=True):
arm='right'
connect(use_gui=viewer)
robot = load_pybullet("franka_panda/panda.urdf", fixed_base=True)
tableUid = load_pybullet("table/table.urdf", fixed_base=True, basePosition=[0.5,0,-0.65])
#workspace = load_pybullet("plane.urdf", fixed_base=True)
set_camera(yaw=0, pitch=-40, distance=1.5, target_position=(0.55, -0.35, 0.2))
cprint('hello Panda! <ctrl+left mouse> to pan', 'green')
wait_for_user()
# create a box to be picked up
# see: https://pybullet-planning.readthedocs.io/en/latest/reference/generated/pybullet_planning.interfaces.env_manager.create_box.html#pybullet_planning.interfaces.env_manager.create_box
block = create_box(0.2, 0.2, 0.4)
block_x = 0.4
block_y = 0.2
block_z = 0.0
set_pose(block, Pose(Point(x=block_x, y=block_y, z=block_z), Euler(yaw=np.pi/2)))
block1 = create_cylinder(0.1, 0.4)
block1_x = 0.8
block1_y = -0.2
block1_z = 0.0
set_pose(block1, Pose(Point(x=block1_x, y=block1_y, z=block1_z), Euler(yaw=np.pi/2)))
ik_joints = get_movable_joints(robot)
ik_joint_names = get_joint_names(robot, ik_joints)
# * if a subset of joints is used, use:
#panda_joints = joints_from_names(robot, ik_joint_names[:7]) # this will disable the gantry-x joint
cprint('Used joints: {}'.format(get_joint_names(robot, ik_joints)), 'yellow')
ee_link = get_link_name(robot, 11)
print(ee_link)
# * get a joint configuration sample function:
# it separately sample each joint value within the feasible range
sample_fn = get_sample_fn(robot, ik_joints)
cprint('Randomly sample robot configuration and set them! (no collision check is performed)', 'blue')
wait_for_user()
for i in range(5):
# randomly sample a joint conf
sampled_conf = sample_fn()
set_joint_positions(robot, ik_joints, sampled_conf)
cprint('#{} | Conf sampeld: {}'.format(i, sampled_conf), 'green')
wait_for_user()
# * now let's plan a trajectory
# we use y-z-arm 6 joint all together here
cprint('Randomly sample robot start/end configuration and comptue a motion plan! (no self-collision check is performed)', 'blue')
print('Disabled collision links needs to be given (can be parsed from a SRDF via compas_fab)')
for _ in range(5):
print('='*10)
'''
q1 = list(sample_fn())
# intentionly make the robot needs to cross the collision object
# let it start from the right side
q1[0] = 0.
q1[1] = 0
'''
#set_joint_positions(robot, ik_joints, q1)
#cprint('Sampled start conf: {}'.format(q1), 'cyan')
#wait_for_user()
# let it ends at the left side
'''
q2 = list(sample_fn())
q2[0] = 0.5
q2[1] = 0.5
cprint('Sampled end conf: {}'.format(q2), 'cyan')
'''
'''
path = plan_joint_motion(robot, ik_joints, q2, obstacles=[block], self_collisions=False,
custom_limits={ik_joints[0]:[0.0, 1.2]})
if path is None:
cprint('no plan found', 'red')
continue
else:
wait_for_user('a motion plan is found! Press enter to start simulating!')
'''
sol = sub_inverse_kinematics(robot, ik_joints[0], 11, ([0.6, 0.0, 0.3], [1,0,0,0]))
print(sol)
# adjusting this number will adjust the simulation speed
time_step = 0.03
for conf in path:
set_joint_positions(robot, ik_joints, conf)
wait_for_duration(time_step)