def add_scales(task, ros_world):
scale_stackings = {}
holding = {grasp.obj_name for grasp in task.init_holding.values()}
with ClientSaver(ros_world.client):
perception = ros_world.perception
items = sorted(set(perception.get_items()) - holding,
key=lambda n: get_point(ros_world.get_body(n))[1],
reverse=False) # Right to left
for i, item in enumerate(items):
if not POUR and (get_type(item) != SCOOP_CUP):
continue
item_body = ros_world.get_body(item)
scale = create_name(SCALE_TYPE, i + 1)
with HideOutput():
scale_body = load_pybullet(get_body_urdf(get_type(scale)),
fixed_base=True)
ros_world.perception.sim_bodies[scale] = scale_body
ros_world.perception.sim_items[scale] = None
item_z = stable_z(item_body, scale_body)
scale_pose_item = Pose(
point=Point(z=-item_z)) # TODO: relies on origin in base
set_pose(scale_body, multiply(get_pose(item_body),
scale_pose_item))
roll, pitch, _ = get_euler(scale_body)
set_euler(scale_body, [roll, pitch, 0])
scale_stackings[scale] = item
#wait_for_user()
return scale_stackings
def mirror_robot(robot1, node_points):
# TODO: place robots side by side or diagonal across
set_extrusion_camera(node_points, theta=-np.pi / 3)
#draw_pose(Pose())
centroid = np.average(node_points, axis=0)
centroid_pose = Pose(point=centroid)
#draw_pose(Pose(point=centroid))
# print(centroid)
scale = 0. # 0.15
vector = get_point(robot1) - centroid
set_pose(robot1, Pose(point=Point(*+scale * vector[:2])))
# Inner product of end-effector z with base->centroid or perpendicular to this line
# Partition by sides
robot2 = load_robot()
set_pose(
robot2,
Pose(point=Point(*-(2 + scale) * vector[:2]), euler=Euler(yaw=np.pi)))
# robots = [robot1]
robots = [robot1, robot2]
for robot in robots:
set_configuration(robot, DUAL_CONF)
# joint1 = get_movable_joints(robot)[0]
# set_joint_position(robot, joint1, np.pi / 8)
draw_pose(get_pose(robot), length=0.25)
return robots
def pour_beads(world,
cup_name,
beads,
reset_contained=False,
fix_outside=True,
cup_thickness=0.01,
bead_offset=0.01,
drop_rate=0.02,
**kwargs):
if not beads:
return set()
start_time = time.time()
# TODO: compute the radius of each bead
bead_radius = np.average(approximate_as_prism(beads[0])) / 2.
masses = list(map(get_mass, beads))
savers = list(map(BodySaver, beads))
for body in beads:
set_mass(body, 0)
cup = world.get_body(cup_name)
local_center, (diameter, height) = approximate_as_cylinder(cup)
center = get_point(cup) + local_center
interior_radius = max(0.0, diameter / 2. - bead_radius - cup_thickness)
# TODO: fill up to a certain threshold
ty = get_type(cup_name)
if ty in SPOON_DIAMETERS:
# TODO: do this in a more principled way
interior_radius = 0
center[1] += (SPOON_LENGTHS[ty] - SPOON_DIAMETERS[ty]) / 2.
# TODO: some sneak out through the bottom
# TODO: could reduce gravity while filling
world.controller.set_gravity()
for i, bead in enumerate(beads):
# TODO: halton sequence
x, y = center[:2] + np.random.uniform(
0, interior_radius) * unit_from_theta(
np.random.uniform(-np.pi, np.pi))
new_z = center[2] + height / 2. + bead_radius + bead_offset
set_point(bead, [x, y, new_z])
set_mass(bead, masses[i])
world.controller.rest_for_duration(drop_rate)
world.controller.rest_for_duration(BEADS_REST)
print('Simulated {} beads in {:3f} seconds'.format(
len(beads), elapsed_time(start_time)))
contained_beads = get_contained_beads(cup, beads, **kwargs)
#wait_for_user()
for body in beads:
if fix_outside and (body not in contained_beads):
set_mass(body, 0)
for saver in savers:
if reset_contained or (saver.body not in contained_beads):
saver.restore()
#wait_for_user()
return contained_beads
def update_world(world, target_body, camera_point=VIEWER_POINT):
pr2 = world.perception.pr2
with ClientSaver(world.perception.client):
open_arm(pr2, LEFT_ARM)
open_arm(pr2, RIGHT_ARM)
set_group_conf(pr2, 'torso', [TORSO_POSITION])
set_group_conf(pr2, arm_from_arm('left'), WIDE_LEFT_ARM)
set_group_conf(pr2, arm_from_arm('right'), rightarm_from_leftarm(WIDE_LEFT_ARM))
target_point = get_point(target_body)
head_conf = inverse_visibility(pr2, target_point, head_name=CAMERA_OPTICAL_FRAME) # Must come after torso
set_group_conf(pr2, 'head', head_conf)
set_camera_pose(camera_point, target_point=target_point)
def draw_push_goal(world, block_name, pos2d):
block_z = get_point(world.get_body(block_name))[2]
handles = draw_point(np.append(pos2d, [block_z]),
size=0.05,
color=(1, 0, 0))
lower, upper = POSE2D_RANGE
handles.extend(
add_segments([
(lower[0], lower[1], block_z),
(upper[0], lower[1], block_z),
(upper[0], upper[1], block_z),
(lower[0], upper[1], block_z),
],
closed=True))
return handles
def _update_custom_limits(self, min_extent=0.0):
#robot_extent = get_aabb_extent(get_aabb(self.robot))
# Scaling by 0.5 to prevent getting caught in corners
#min_extent = 0.5 * min(robot_extent[:2]) * np.ones(2) / 2
world_aabb = self.get_world_aabb()
full_lower, full_upper = world_aabb
base_limits = (full_lower[:2] - min_extent,
full_upper[:2] + min_extent)
base_limits[1][0] = COMPUTER_X - min_extent # TODO: robot radius
base_limits[0][1] += 0.1
base_limits[1][1] -= 0.1
for handle in self.base_limits_handles:
remove_debug(handle)
self.base_limits_handles = []
#self.base_limits_handles.extend(draw_aabb(world_aabb))
z = get_point(self.floor)[2] + 1e-2
self.base_limits_handles.extend(draw_base_limits(base_limits, z=z))
self.custom_limits = custom_limits_from_base_limits(
self.robot, base_limits)
return self.custom_limits
def _initialize_environment(self):
# wall to fridge: 4cm
# fridge to goal: 1.5cm
# hitman to range: 3.5cm
# range to indigo: 3.5cm
self.environment_poses = read_json(POSES_PATH)
root_from_world = get_link_pose(self.kitchen, self.world_link)
for name, world_from_part in self.environment_poses.items():
if name in ['range']:
continue
visual_path = os.path.join(KITCHEN_LEFT_PATH,
'{}.obj'.format(name))
collision_path = os.path.join(KITCHEN_LEFT_PATH,
'{}_collision.obj'.format(name))
mesh_path = None
for path in [collision_path, visual_path]:
if os.path.exists(path):
mesh_path = path
break
if mesh_path is None:
continue
body = load_pybullet(mesh_path, fixed_base=True)
root_from_part = multiply(root_from_world, world_from_part)
if name in ['axe', 'dishwasher', 'echo', 'fox', 'golf']:
(pos, quat) = root_from_part
# TODO: still not totally aligned
pos = np.array(pos) + np.array([0, -0.035, 0]) # , -0.005])
root_from_part = (pos, quat)
self.environment_bodies[name] = body
set_pose(body, root_from_part)
# TODO: release bounding box or convex hull
# TODO: static object nonconvex collisions
if TABLE_NAME in self.environment_bodies:
body = self.environment_bodies[TABLE_NAME]
_, (w, l, _) = approximate_as_prism(body)
_, _, z = get_point(body)
new_pose = Pose(Point(TABLE_X + l / 2, -TABLE_Y, z),
Euler(yaw=np.pi / 2))
set_pose(body, new_pose)
def __init__(self,
robot_name=FRANKA_CARTER,
use_gui=True,
full_kitchen=False):
self.task = None
self.interface = None
self.client = connect(use_gui=use_gui)
set_real_time(False)
#set_caching(False) # Seems to make things worse
disable_gravity()
add_data_path()
set_camera_pose(camera_point=[2, -1.5, 1])
if DEBUG:
draw_pose(Pose(), length=3)
#self.kitchen_yaml = load_yaml(KITCHEN_YAML)
with HideOutput(enable=True):
self.kitchen = load_pybullet(KITCHEN_PATH,
fixed_base=True,
cylinder=True)
self.floor = load_pybullet('plane.urdf', fixed_base=True)
z = stable_z(self.kitchen, self.floor) - get_point(self.floor)[2]
point = np.array(get_point(self.kitchen)) - np.array([0, 0, z])
set_point(self.floor, point)
self.robot_name = robot_name
if self.robot_name == FRANKA_CARTER:
urdf_path, yaml_path = FRANKA_CARTER_PATH, None
#urdf_path, yaml_path = FRANKA_CARTER_PATH, FRANKA_YAML
#elif self.robot_name == EVE:
# urdf_path, yaml_path = EVE_PATH, None
else:
raise ValueError(self.robot_name)
self.robot_yaml = yaml_path if yaml_path is None else load_yaml(
yaml_path)
with HideOutput(enable=True):
self.robot = load_pybullet(urdf_path)
#dump_body(self.robot)
#chassis_pose = get_link_pose(self.robot, link_from_name(self.robot, 'chassis_link'))
#wheel_pose = get_link_pose(self.robot, link_from_name(self.robot, 'left_wheel_link'))
#wait_for_user()
set_point(self.robot, Point(z=stable_z(self.robot, self.floor)))
self.set_initial_conf()
self.gripper = create_gripper(self.robot)
self.environment_bodies = {}
if full_kitchen:
self._initialize_environment()
self._initialize_ik(urdf_path)
self.initial_saver = WorldSaver()
self.body_from_name = {}
# self.path_from_name = {}
self.names_from_type = {}
self.custom_limits = {}
self.base_limits_handles = []
self.cameras = {}
self.disabled_collisions = set()
if self.robot_name == FRANKA_CARTER:
self.disabled_collisions.update(
tuple(link_from_name(self.robot, link) for link in pair)
for pair in DISABLED_FRANKA_COLLISIONS)
self.carry_conf = FConf(self.robot, self.arm_joints, self.default_conf)
#self.calibrate_conf = Conf(self.robot, self.arm_joints, load_calibrate_conf(side='left'))
self.calibrate_conf = FConf(
self.robot, self.arm_joints,
self.default_conf) # Must differ from carry_conf
self.special_confs = [self.carry_conf] #, self.calibrate_conf]
self.open_gq = FConf(self.robot, self.gripper_joints,
get_max_limits(self.robot, self.gripper_joints))
self.closed_gq = FConf(self.robot, self.gripper_joints,
get_min_limits(self.robot, self.gripper_joints))
self.gripper_confs = [self.open_gq, self.closed_gq]
self.open_kitchen_confs = {
joint: FConf(self.kitchen, [joint], [self.open_conf(joint)])
for joint in self.kitchen_joints
}
self.closed_kitchen_confs = {
joint: FConf(self.kitchen, [joint], [self.closed_conf(joint)])
for joint in self.kitchen_joints
}
self._update_custom_limits()
self._update_initial()
def main(floor_width=2.0):
# Creates a pybullet world and a visualizer for it
connect(use_gui=True)
identity_pose = (unit_point(), unit_quat())
origin_handles = draw_pose(
identity_pose, length=1.0
) # Draws the origin coordinate system (x:RED, y:GREEN, z:BLUE)
# Bodies are described by an integer index
floor = create_box(w=floor_width, l=floor_width, h=0.001,
color=TAN) # Creates a tan box object for the floor
set_point(floor,
[0, 0, -0.001 / 2.]) # Sets the [x,y,z] translation of the floor
obstacle = create_box(w=0.5, l=0.5, h=0.1,
color=RED) # Creates a red box obstacle
set_point(
obstacle,
[0.5, 0.5, 0.1 / 2.]) # Sets the [x,y,z] position of the obstacle
print('Position:', get_point(obstacle))
set_euler(obstacle,
[0, 0, np.pi / 4
]) # Sets the [roll,pitch,yaw] orientation of the obstacle
print('Orientation:', get_euler(obstacle))
with LockRenderer(
): # Temporarily prevents the renderer from updating for improved loading efficiency
with HideOutput(): # Temporarily suppresses pybullet output
robot = load_model(ROOMBA_URDF) # Loads a robot from a *.urdf file
robot_z = stable_z(
robot, floor
) # Returns the z offset required for robot to be placed on floor
set_point(robot,
[0, 0, robot_z]) # Sets the z position of the robot
dump_body(robot) # Prints joint and link information about robot
set_all_static()
# Joints are also described by an integer index
# The turtlebot has explicit joints representing x, y, theta
x_joint = joint_from_name(robot, 'x') # Looks up the robot joint named 'x'
y_joint = joint_from_name(robot, 'y') # Looks up the robot joint named 'y'
theta_joint = joint_from_name(
robot, 'theta') # Looks up the robot joint named 'theta'
joints = [x_joint, y_joint, theta_joint]
base_link = link_from_name(
robot, 'base_link') # Looks up the robot link named 'base_link'
world_from_obstacle = get_pose(
obstacle
) # Returns the pose of the origin of obstacle wrt the world frame
obstacle_aabb = get_subtree_aabb(obstacle)
draw_aabb(obstacle_aabb)
random.seed(0) # Sets the random number generator state
handles = []
for i in range(10):
for handle in handles:
remove_debug(handle)
print('\nIteration: {}'.format(i))
x = random.uniform(-floor_width / 2., floor_width / 2.)
set_joint_position(robot, x_joint,
x) # Sets the current value of the x joint
y = random.uniform(-floor_width / 2., floor_width / 2.)
set_joint_position(robot, y_joint,
y) # Sets the current value of the y joint
yaw = random.uniform(-np.pi, np.pi)
set_joint_position(robot, theta_joint,
yaw) # Sets the current value of the theta joint
values = get_joint_positions(
robot,
joints) # Obtains the current values for the specified joints
print('Joint values: [x={:.3f}, y={:.3f}, yaw={:.3f}]'.format(*values))
world_from_robot = get_link_pose(
robot,
base_link) # Returns the pose of base_link wrt the world frame
position, quaternion = world_from_robot # Decomposing pose into position and orientation (quaternion)
x, y, z = position # Decomposing position into x, y, z
print('Base link position: [x={:.3f}, y={:.3f}, z={:.3f}]'.format(
x, y, z))
euler = euler_from_quat(
quaternion) # Converting from quaternion to euler angles
roll, pitch, yaw = euler # Decomposing orientation into roll, pitch, yaw
print('Base link orientation: [roll={:.3f}, pitch={:.3f}, yaw={:.3f}]'.
format(roll, pitch, yaw))
handles.extend(
draw_pose(world_from_robot, length=0.5)
) # # Draws the base coordinate system (x:RED, y:GREEN, z:BLUE)
obstacle_from_robot = multiply(
invert(world_from_obstacle),
world_from_robot) # Relative transformation from robot to obstacle
robot_aabb = get_subtree_aabb(
robot,
base_link) # Computes the robot's axis-aligned bounding box (AABB)
lower, upper = robot_aabb # Decomposing the AABB into the lower and upper extrema
center = (lower + upper) / 2. # Computing the center of the AABB
extent = upper - lower # Computing the dimensions of the AABB
handles.extend(draw_aabb(robot_aabb))
collision = pairwise_collision(
robot, obstacle
) # Checks whether robot is currently colliding with obstacle
print('Collision: {}'.format(collision))
wait_for_duration(1.0) # Like sleep() but also updates the viewer
wait_for_user() # Like raw_input() but also updates the viewer
# Destroys the pybullet world
disconnect()
def eval_task_with_seed(domain,
seed,
learner,
sample_strategy=DIVERSELK,
task_lengthscale=None,
obstacle=True):
if task_lengthscale is not None:
learner.task_lengthscale = task_lengthscale
print('sample a new task')
current_wd, trial_wd = start_task()
### fill in additional object
item_ranges = {}
###
sim_world, collector, task, feature, evalfunc, saver = sample_task_with_seed(
domain.skill, seed=seed, visualize=False, item_ranges=item_ranges)
context = domain.context_from_feature(feature)
print('context=', *context)
# create coffee machine
if obstacle:
bowl_name = 'bowl'
cup_name = 'cup'
if domain.skill == 'scoop':
cup_name = 'spoon'
for key in sim_world.perception.sim_bodies:
if bowl_name in key:
bowl_name = key
if cup_name in key:
cup_name = key
cylinder_size = (.01, .03)
dim_bowl = get_aabb_extent(
p.getAABB(sim_world.perception.sim_bodies[bowl_name]))
dim_cup = get_aabb_extent(
p.getAABB(sim_world.perception.sim_bodies[cup_name]))
bowl_point = get_point(sim_world.perception.sim_bodies[bowl_name])
bowl_point = np.array(bowl_point)
cylinder_point = bowl_point.copy()
cylinder_offset = 1.2
if domain.skill == 'scoop':
cylinder_offset = 1.6
cylinder_point[2] += (dim_bowl[2] + dim_cup[2]) * (np.random.random_sample() * .4 + cylinder_offset) \
+ cylinder_size[1] / 2.
# TODO figure out the task space
cylinder_name = create_name('cylinder', 1)
obstacle = create_cylinder(*cylinder_size, color=(0, 0, 0, 1))
INFO_FROM_BODY[(CLIENT, obstacle)] = ModelInfo(cylinder_name,
cylinder_size, 1, 1)
sim_world.perception.sim_bodies[cylinder_name] = obstacle
sim_world.perception.sim_items[cylinder_name] = None
set_pose(obstacle, (cylinder_point, unit_quat()))
box_name = create_name('box', 1)
box1_size = (max(.17,
dim_bowl[0] / 2 + cylinder_size[0] * 2), 0.01, 0.01)
if domain.skill == 'scoop':
box1_size = (max(.23,
dim_bowl[0] / 2 + cylinder_size[0] * 2 + 0.05),
0.01, 0.01)
obstacle = create_box(*box1_size)
INFO_FROM_BODY[(CLIENT, obstacle)] = ModelInfo(box_name, box1_size, 1,
1)
sim_world.perception.sim_bodies[box_name] = obstacle
sim_world.perception.sim_items[box_name] = None
box1_point = cylinder_point.copy()
box1_point[2] += cylinder_size[1] / 2 + box1_size[2] / 2
box1_point[0] += box1_size[0] / 2 - cylinder_size[0] * 2
set_pose(obstacle, (box1_point, unit_quat()))
box_name = create_name('box', 2)
box2_size = (0.01, .01,
box1_point[2] - bowl_point[2] + box1_size[2] / 2)
obstacle = create_box(*box2_size)
INFO_FROM_BODY[(CLIENT, obstacle)] = ModelInfo(box_name, box2_size, 1,
1)
sim_world.perception.sim_bodies[box_name] = obstacle
sim_world.perception.sim_items[box_name] = None
box2_point = bowl_point.copy()
box2_point[2] += box2_size[2] / 2
box2_point[0] = box1_point[0] + box1_size[0] / 2
set_pose(obstacle, (box2_point, unit_quat()))
result = get_sample_strategy_result(sample_strategy, learner, context,
sim_world, collector, task, feature,
evalfunc, saver)
print('context=', *context)
complete_task(sim_world, current_wd, trial_wd)
return result
def get_floor_z(world, floor_z=0.005):
return get_point(world.floor)[2] + floor_z
