def main(time_step=0.01):
parser = create_parser()
parser.add_argument('-teleport',
action='store_true',
help='Teleports between configurations')
parser.add_argument('-viewer',
action='store_true',
help='enable the viewer while planning')
# TODO: argument for selecting prior
args = parser.parse_args()
print('Arguments:', args)
# TODO: nonuniform distribution to bias towards other actions
# TODO: closed world and open world
real_world = connect(use_gui=not args.viewer)
add_data_path()
task, state = get_problem1(localized='rooms',
p_other=0.25) # surfaces | rooms
for body in task.get_bodies():
add_body_name(body)
robot = task.robot
#dump_body(robot)
assert (USE_DRAKE_PR2 == is_drake_pr2(robot))
attach_viewcone(robot) # Doesn't work for the normal pr2?
draw_base_limits(get_base_limits(robot), color=(0, 1, 0))
#wait_for_user()
# TODO: partially observable values
# TODO: base movements preventing pick without look
# TODO: do everything in local coordinate frame
# TODO: automatically determine an action/command cannot be applied
# TODO: convert numpy arrays into what they are close to
# TODO: compute whether a plan will still achieve a goal and do that
# TODO: update the initial state directly and then just redraw it to ensure uniqueness
step = 0
while True:
step += 1
print('\n' + 50 * '-')
print(step, state)
wait_for_user()
#print({b: p.value for b, p in state.poses.items()})
with ClientSaver():
commands = plan_commands(state, args)
print()
if commands is None:
print('Failure!')
break
if not commands:
print('Success!')
break
apply_commands(state, commands, time_step=time_step)
print(state)
wait_for_user()
disconnect()
def packed(arm='left', grasp_type='top', num=5):
# TODO: packing problem where you have to place in one direction
base_extent = 5.0
base_limits = (-base_extent / 2. * np.ones(2),
base_extent / 2. * np.ones(2))
block_width = 0.07
block_height = 0.1
#block_height = 2*block_width
block_area = block_width * block_width
#plate_width = 2*math.sqrt(num*block_area)
plate_width = 0.27
#plate_width = 0.28
#plate_width = 0.3
print('Width:', plate_width)
plate_width = min(plate_width, 0.6)
plate_height = 0.001
other_arm = get_other_arm(arm)
initial_conf = get_carry_conf(arm, grasp_type)
add_data_path()
floor = load_pybullet("plane.urdf")
pr2 = create_pr2()
set_arm_conf(pr2, arm, initial_conf)
open_arm(pr2, arm)
set_arm_conf(pr2, other_arm, arm_conf(other_arm, REST_LEFT_ARM))
close_arm(pr2, other_arm)
set_group_conf(pr2, 'base',
[-1.0, 0, 0]) # Be careful to not set the pr2's pose
table = create_table()
plate = create_box(plate_width, plate_width, plate_height, color=GREEN)
plate_z = stable_z(plate, table)
set_point(plate, Point(z=plate_z))
surfaces = [table, plate]
blocks = [
create_box(block_width, block_width, block_height, color=BLUE)
for _ in range(num)
]
initial_surfaces = {block: table for block in blocks}
min_distances = {block: 0.05 for block in blocks}
sample_placements(initial_surfaces, min_distances=min_distances)
return Problem(
robot=pr2,
movable=blocks,
arms=[arm],
grasp_types=[grasp_type],
surfaces=surfaces,
#goal_holding=[(arm, block) for block in blocks])
goal_on=[(block, plate) for block in blocks],
base_limits=base_limits)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('-viewer', action='store_true', help='enable the viewer while planning')
args = parser.parse_args()
print(args)
connect(use_gui=True)
with LockRenderer():
draw_pose(unit_pose(), length=1)
floor = create_floor()
with HideOutput():
robot = load_pybullet(get_model_path(ROOMBA_URDF), fixed_base=True, scale=2.0)
for link in get_all_links(robot):
set_color(robot, link=link, color=ORANGE)
robot_z = stable_z(robot, floor)
set_point(robot, Point(z=robot_z))
#set_base_conf(robot, rover_confs[i])
data_path = add_data_path()
shelf, = load_model(os.path.join(data_path, KIVA_SHELF_SDF), fixed_base=True) # TODO: returns a tuple
dump_body(shelf)
#draw_aabb(get_aabb(shelf))
wait_for_user()
disconnect()
def main(time_step=0.01):
# TODO: closed world and open world
real_world = connect(use_gui=True)
add_data_path()
task, state = get_problem1(localized='rooms',
p_other=0.5) # surfaces | rooms
for body in task.get_bodies():
add_body_name(body)
robot = task.robot
#dump_body(robot)
assert (USE_DRAKE_PR2 == is_drake_pr2(robot))
attach_viewcone(robot) # Doesn't work for the normal pr2?
draw_base_limits(get_base_limits(robot), color=(0, 1, 0))
#wait_for_interrupt()
# TODO: partially observable values
# TODO: base movements preventing pick without look
# TODO: do everything in local coordinate frame
# TODO: automatically determine an action/command cannot be applied
# TODO: convert numpy arrays into what they are close to
# TODO: compute whether a plan will still achieve a goal and do that
# TODO: update the initial state directly and then just redraw it to ensure uniqueness
step = 0
while True:
step += 1
print('\n' + 50 * '-')
print(step, state)
#print({b: p.value for b, p in state.poses.items()})
with ClientSaver():
commands = plan_commands(state)
print()
if commands is None:
print('Failure!')
break
if not commands:
print('Success!')
break
apply_commands(state, commands, time_step=time_step)
print(state)
wait_for_interrupt()
disconnect()
def problem1(n_rovers=1, n_objectives=1, n_rocks=2, n_soil=2, n_stores=1):
base_extent = 5.0
base_limits = (-base_extent / 2. * np.ones(2),
base_extent / 2. * np.ones(2))
floor = create_box(base_extent, base_extent, 0.001,
color=TAN) # TODO: two rooms
set_point(floor, Point(z=-0.001 / 2.))
add_data_path()
lander = load_pybullet(HUSKY_URDF, scale=1)
lander_z = stable_z(lander, floor)
set_point(lander, Point(-2, -2, lander_z))
#wait_for_user()
mount_width = 0.5
mound_height = mount_width
mound1 = create_box(mount_width, mount_width, mound_height, color=GREY)
set_point(mound1, [+2, 1.5, mound_height / 2.])
mound2 = create_box(mount_width, mount_width, mound_height, color=GREY)
set_point(mound2, [-2, 1.5, mound_height / 2.])
initial_surfaces = {}
rover_confs = [(+1, -1.75, np.pi), (-1, -1.75, 0)]
assert n_rovers <= len(rover_confs)
#body_names = map(get_name, env.GetBodies())
landers = [lander]
stores = ['store{}'.format(i) for i in range(n_stores)]
#affine_limits = aabb_extrema(aabb_union([aabb_from_body(body) for body in env.GetBodies()])).T
rovers = []
for i in range(n_rovers):
robot = create_pr2()
dump_body(robot) # camera_rgb_optical_frame
rovers.append(robot)
set_group_conf(robot, 'base', rover_confs[i])
for arm in ARM_NAMES:
set_arm_conf(robot, arm, arm_conf(arm, REST_LEFT_ARM))
close_arm(robot, arm)
obj_width = 0.07
obj_height = 0.2
objectives = []
for _ in range(n_objectives):
body = create_box(obj_width, obj_width, obj_height, color=BLUE)
objectives.append(body)
initial_surfaces[body] = mound1
for _ in range(n_objectives):
body = create_box(obj_width, obj_width, obj_height, color=RED)
initial_surfaces[body] = mound2
# TODO: it is moving to intermediate locations attempting to reach the rocks
rocks = []
for _ in range(n_rocks):
body = create_box(0.075, 0.075, 0.05, color=BLACK)
rocks.append(body)
initial_surfaces[body] = floor
soils = []
for _ in range(n_soil):
body = create_box(0.1, 0.1, 0.025, color=BROWN)
soils.append(body)
initial_surfaces[body] = floor
sample_placements(initial_surfaces)
#for name in rocks + soils:
# env.GetKinBody(name).Enable(False)
return RoversProblem(rovers, landers, objectives, rocks, soils, stores,
base_limits)
def rovers1(n_rovers=2,
n_objectives=4,
n_rocks=3,
n_soil=3,
n_stores=1,
n_obstacles=8):
base_extent = 5.0
base_limits = (-base_extent / 2. * np.ones(2),
base_extent / 2. * np.ones(2))
mount_width = 0.5
mound_height = 0.1
floor = create_box(base_extent, base_extent, 0.001,
color=TAN) # TODO: two rooms
set_point(floor, Point(z=-0.001 / 2.))
wall1 = create_box(base_extent + mound_height,
mound_height,
mound_height,
color=GREY)
set_point(wall1, Point(y=base_extent / 2., z=mound_height / 2.))
wall2 = create_box(base_extent + mound_height,
mound_height,
mound_height,
color=GREY)
set_point(wall2, Point(y=-base_extent / 2., z=mound_height / 2.))
wall3 = create_box(mound_height,
base_extent + mound_height,
mound_height,
color=GREY)
set_point(wall3, Point(x=base_extent / 2., z=mound_height / 2.))
wall4 = create_box(mound_height,
base_extent + mound_height,
mound_height,
color=GREY)
set_point(wall4, Point(x=-base_extent / 2., z=mound_height / 2.))
# TODO: can add obstacles along the wall
wall = create_box(mound_height, base_extent, mound_height,
color=GREY) # TODO: two rooms
set_point(wall, Point(z=mound_height / 2.))
add_data_path()
with HideOutput():
lander = load_pybullet(HUSKY_URDF, scale=1)
lander_z = stable_z(lander, floor)
set_point(lander, Point(-1.9, -2, lander_z))
mound1 = create_box(mount_width, mount_width, mound_height, color=GREY)
set_point(mound1, [+2, 2, mound_height / 2.])
mound2 = create_box(mount_width, mount_width, mound_height, color=GREY)
set_point(mound2, [-2, 2, mound_height / 2.])
mound3 = create_box(mount_width, mount_width, mound_height, color=GREY)
set_point(mound3, [+0.5, 2, mound_height / 2.])
mound4 = create_box(mount_width, mount_width, mound_height, color=GREY)
set_point(mound4, [-0.5, 2, mound_height / 2.])
mounds = [mound1, mound2, mound3, mound4]
random.shuffle(mounds)
body_types = []
initial_surfaces = OrderedDict()
min_distances = {}
for _ in range(n_obstacles):
body = create_box(mound_height,
mound_height,
4 * mound_height,
color=GREY)
initial_surfaces[body] = floor
rover_confs = [(+1, -1.75, np.pi), (-1, -1.75, 0)]
assert n_rovers <= len(rover_confs)
landers = [lander]
stores = ['store{}'.format(i) for i in range(n_stores)]
rovers = []
for i in range(n_rovers):
# camera_rgb_optical_frame
with HideOutput():
rover = load_model(TURTLEBOT_URDF)
robot_z = stable_z(rover, floor)
set_point(rover, Point(z=robot_z))
#handles = draw_aabb(get_aabb(rover)) # Includes the origin
#print(get_center_extent(rover))
#wait_for_user()
set_base_conf(rover, rover_confs[i])
rovers.append(rover)
#dump_body(rover)
#draw_pose(get_link_pose(rover, link_from_name(rover, KINECT_FRAME)))
obj_width = 0.07
obj_height = 0.2
objectives = []
for i in range(n_objectives):
body = create_box(obj_width, obj_width, obj_height, color=BLUE)
objectives.append(body)
#initial_surfaces[body] = random.choice(mounds)
initial_surfaces[body] = mounds[i]
min_distances.update({r: 0.05 for r in objectives})
# TODO: it is moving to intermediate locations attempting to reach the rocks
rocks = []
for _ in range(n_rocks):
body = create_box(0.075, 0.075, 0.01, color=BLACK)
rocks.append(body)
body_types.append((body, 'stone'))
for _ in range(n_soil):
body = create_box(0.1, 0.1, 0.005, color=BROWN)
rocks.append(body)
body_types.append((body, 'soil'))
soils = [] # Treating soil as rocks for simplicity
initial_surfaces.update({r: floor for r in rocks})
min_distances.update({r: 0.2 for r in rocks})
sample_placements(initial_surfaces, min_distances=min_distances)
return RoversProblem(rovers, landers, objectives, rocks, soils, stores,
base_limits, body_types)
def blocked(arm='left', grasp_type='side', num=1):
x_extent = 10.0
base_limits = (-x_extent / 2. * np.ones(2), x_extent / 2. * np.ones(2))
block_width = 0.07
#block_height = 0.1
block_height = 2 * block_width
#block_height = 0.2
plate_height = 0.001
table_x = (x_extent - 1) / 2.
other_arm = get_other_arm(arm)
initial_conf = get_carry_conf(arm, grasp_type)
add_data_path()
floor = load_pybullet("plane.urdf")
pr2 = create_pr2()
set_arm_conf(pr2, arm, initial_conf)
open_arm(pr2, arm)
set_arm_conf(pr2, other_arm, arm_conf(other_arm, REST_LEFT_ARM))
close_arm(pr2, other_arm)
set_group_conf(
pr2, 'base',
[x_extent / 4, 0, 0]) # Be careful to not set the pr2's pose
table1 = create_table()
set_point(table1, Point(x=+table_x, y=0))
table2 = create_table()
set_point(table2, Point(x=-table_x, y=0))
#table3 = create_table()
#set_point(table3, Point(x=0, y=0))
plate = create_box(0.6, 0.6, plate_height, color=GREEN)
x, y, _ = get_point(table1)
plate_z = stable_z(plate, table1)
set_point(plate, Point(x=x, y=y - 0.3, z=plate_z))
#surfaces = [table1, table2, table3, plate]
surfaces = [table1, table2, plate]
green1 = create_box(block_width, block_width, block_height, color=BLUE)
green1_z = stable_z(green1, table1)
set_point(green1, Point(x=x, y=y + 0.3, z=green1_z))
# TODO: can consider a fixed wall here instead
spacing = 0.15
#red_directions = [(-1, 0), (+1, 0), (0, -1), (0, +1)]
red_directions = [(-1, 0)]
#red_directions = []
red_bodies = []
for red_direction in red_directions:
red = create_box(block_width, block_width, block_height, color=RED)
red_bodies.append(red)
x, y = get_point(green1)[:2] + spacing * np.array(red_direction)
z = stable_z(red, table1)
set_point(red, Point(x=x, y=y, z=z))
wall1 = create_box(0.01, 2 * spacing, block_height, color=GREY)
wall2 = create_box(spacing, 0.01, block_height, color=GREY)
wall3 = create_box(spacing, 0.01, block_height, color=GREY)
z = stable_z(wall1, table1)
x, y = get_point(green1)[:2]
set_point(wall1, Point(x=x + spacing, y=y, z=z))
set_point(wall2, Point(x=x + spacing / 2, y=y + spacing, z=z))
set_point(wall3, Point(x=x + spacing / 2, y=y - spacing, z=z))
green_bodies = [
create_box(block_width, block_width, block_height, color=BLUE)
for _ in range(num)
]
body_types = [(b, 'green')
for b in [green1] + green_bodies] # + [(table1, 'sink')]
movable = [green1] + green_bodies + red_bodies
initial_surfaces = {block: table2 for block in green_bodies}
sample_placements(initial_surfaces)
return Problem(
robot=pr2,
movable=movable,
arms=[arm],
grasp_types=[grasp_type],
surfaces=surfaces,
#sinks=[table1],
#goal_holding=[(arm, '?green')],
#goal_cleaned=['?green'],
goal_on=[('?green', plate)],
body_types=body_types,
base_limits=base_limits,
costs=True)