def test_sensor_noise_robot(self):
obs_config = ObservationConfig(
joint_velocities_noise=GaussianNoise(0.01),
joint_forces=False,
task_low_dim_state=False)
scene = Scene(self.pyrep, self.robot, obs_config)
scene.load(ReachTarget(self.pyrep, self.robot))
obs1 = scene.get_observation()
obs2 = scene.get_observation()
self.assertTrue(
np.array_equal(obs1.joint_positions, obs2.joint_positions))
self.assertFalse(
np.array_equal(obs1.joint_velocities, obs2.joint_velocities))
def test_sensor_noise_images(self):
cam_config = CameraConfig(rgb_noise=GaussianNoise(0.05, (.0, 1.)))
obs_config = ObservationConfig(left_shoulder_camera=cam_config,
joint_forces=False,
task_low_dim_state=False)
scene = Scene(self.pyrep, self.robot, obs_config)
scene.load(ReachTarget(self.pyrep, self.robot))
obs1 = scene.get_observation()
obs2 = scene.get_observation()
self.assertTrue(
np.array_equal(obs1.right_shoulder_rgb, obs2.right_shoulder_rgb))
self.assertFalse(
np.array_equal(obs1.left_shoulder_rgb, obs2.left_shoulder_rgb))
self.assertTrue(obs1.left_shoulder_rgb.max() <= 1.0)
self.assertTrue(obs1.left_shoulder_rgb.min() >= 0.0)
def launch(self):
if self._pyrep is not None:
raise RuntimeError('Already called launch!')
self._pyrep = PyRep()
self._pyrep.launch(join(DIR_PATH, TTT_FILE), headless=self._headless)
arm_class, gripper_class, _ = SUPPORTED_ROBOTS[
self._robot_configuration]
# We assume the panda is already loaded in the scene.
if self._robot_configuration is not 'panda':
# Remove the panda from the scene
panda_arm = Panda()
panda_pos = panda_arm.get_position()
panda_arm.remove()
arm_path = join(DIR_PATH, 'robot_ttms',
self._robot_configuration + '.ttm')
self._pyrep.import_model(arm_path)
arm, gripper = arm_class(), gripper_class()
arm.set_position(panda_pos)
else:
arm, gripper = arm_class(), gripper_class()
self._robot = Robot(arm, gripper)
if self._randomize_every is None:
self._scene = Scene(self._pyrep, self._robot, self._obs_config)
else:
self._scene = DomainRandomizationScene(
self._pyrep, self._robot, self._obs_config,
self._randomize_every, self._frequency,
self._visual_randomization_config,
self._dynamics_randomization_config)
self._set_arm_control_action()
def test_run_task_validator(self):
for task_file in TASKS:
test_name = task_file.split('.py')[0]
with self.subTest(task=test_name):
if test_name in FLAKY_TASKS:
self.skipTest('Flaky task.')
sim = PyRep()
ttt_file = os.path.join(
DIR_PATH, '..', '..', 'rlbench', TTT_FILE)
sim.launch(ttt_file, headless=True)
sim.step_ui()
sim.set_simulation_timestep(50.0)
sim.step_ui()
sim.start()
robot = Robot(Panda(), PandaGripper())
obs = ObservationConfig()
obs.set_all(False)
scene = Scene(sim, robot, obs)
sim.start()
task_class = task_file_to_task_class(task_file)
active_task = task_class(sim, robot)
try:
task_smoke(active_task, scene, variation=-1,
max_variations=2, success=0.25)
except Exception as e:
sim.stop()
sim.shutdown()
raise e
sim.stop()
sim.shutdown()
def launch(self):
if self._pyrep is not None:
raise RuntimeError('Already called launch!')
self._pyrep = PyRep()
self._pyrep.launch(join(DIR_PATH, TTT_FILE), headless=self._headless)
self._pyrep.set_simulation_timestep(0.005)
self._robot = Robot(Panda(), PandaGripper())
self._scene = Scene(self._pyrep, self._robot, self._obs_config)
self._set_arm_control_action()
def init(self, display=False):
if self._pyrep is not None:
self.finalize()
with suppress_std_out_and_err():
self._pyrep = PyRep()
# TODO: TTT_FILE should be defined by robot, but now robot depends on launched pyrep
self._pyrep.launch(join(DIR_PATH, TTT_FILE), headless=not display)
self._pyrep.set_simulation_timestep(0.005)
# TODO: Load arm and gripper from name
self._robot = Robot(Panda(), PandaGripper())
self._scene = Scene(self._pyrep, self._robot, self._obs_config)
self._set_arm_control_action()
# Str comparison because sometimes class comparison doesn't work.
if self._task is not None:
self._task.unload()
self._task = self._get_class_by_name(self._task_name, tasks)(self._pyrep, self._robot)
self._scene.load(self._task)
self._pyrep.start()
def launch(self):
if self._pyrep is not None:
raise RuntimeError('Already called launch!')
self._pyrep = PyRep()
self._pyrep.launch(join(DIR_PATH, TTT_FILE), headless=self._headless)
self._pyrep.set_simulation_timestep(0.005)
snake_robot_class, camera_class = SUPPORTED_ROBOTS[self._robot_configuration]
# We assume the panda is already loaded in the scene.
if self._robot_configuration != 'rattler':
raise NotImplementedError("Not implemented the robot")
else:
snake_robot, camera = snake_robot_class(), camera_class()
self._robot = Robot(snake_robot, camera)
self._scene = Scene(self._pyrep, self._robot, self._obs_config)
self._set_control_action()
def task_smoke(task: Task,
scene: Scene,
variation=-1,
demos=4,
success=0.50,
max_variations=3,
test_demos=True):
# -1 variations for all.
print('Running task validator on task: %s' % task.get_name())
# Loading
scene.load(task)
# Number of variations
variation_count = task.variation_count()
if variation_count < 0:
raise TaskValidationError(
"The method 'variation_count' should return a number > 0.")
if variation_count > MAX_VARIATIONS:
raise TaskValidationError(
"This task had %d variations. Currently the limit is set to %d" %
(variation_count, MAX_VARIATIONS))
# Base rotation bounds
base_pos, base_ori = task.base_rotation_bounds()
if len(base_pos) != 3 or len(base_ori) != 3:
raise TaskValidationError(
"The method 'base_rotation_bounds' should return a tuple "
"containing a list of floats.")
# Boundary root
root = task.boundary_root()
if not root.still_exists():
raise TaskValidationError(
"The method 'boundary_root' should return a Dummy that is the root "
"of the task.")
def variation_smoke(i):
print('Running task validator on variation: %d' % i)
attempt_result = False
failed_demos = 0
for j in range(DEMO_ATTEMPTS):
failed_demos = run_demos(i)
attempt_result = (failed_demos / float(demos) <= 1. - success)
if attempt_result:
break
else:
print('Failed on attempt %d. Trying again...' % j)
# Make sure we don't fail too often
if not attempt_result:
raise TaskValidationError(
"Too many failed demo runs. %d of %d demos failed." %
(failed_demos, demos))
else:
print('Variation %d of task %s is good!' % (i, task.get_name()))
if test_demos:
print('%d of %d demos were successful.' %
(demos - failed_demos, demos))
def run_demos(variation_num):
fails = 0
for dr in range(demos):
try:
scene.reset()
desc = scene.init_episode(variation_num, max_attempts=10)
if not isinstance(desc, list) or len(desc) <= 0:
raise TaskValidationError(
"The method 'init_variation' should return a list of "
"string descriptions.")
if test_demos:
demo = scene.get_demo(record=True)
assert len(demo) > 0
except DemoError as e:
fails += 1
print(e)
except Exception as e:
# TODO: check that we don't fall through all of these cases
fails += 1
print(e)
return fails
variations_to_test = [variation]
if variation < 0:
variations_to_test = list(
range(np.minimum(variation_count, max_variations)))
# Task set-up
scene.init_task()
[variation_smoke(i) for i in variations_to_test]
args = parser.parse_args()
python_file = os.path.join(TASKS_PATH, args.task)
if not os.path.isfile(python_file):
raise RuntimeError('Could not find the task file: %s' % python_file)
task_class = task_file_to_task_class(args.task)
DIR_PATH = os.path.dirname(os.path.abspath(__file__))
sim = PyRep()
ttt_file = os.path.join(DIR_PATH, '..', 'rlbench', TTT_FILE)
sim.launch(ttt_file, headless=True)
sim.step_ui()
sim.set_simulation_timestep(0.005)
sim.step_ui()
sim.start()
robot = Robot(Panda(), PandaGripper())
active_task = task_class(sim, robot)
obs = ObservationConfig()
obs.set_all(False)
scene = Scene(sim, robot, obs)
try:
task_smoke(active_task, scene, variation=2)
except TaskValidationError as e:
sim.shutdown()
raise e
sim.shutdown()
print('Validation successful!')
pr.launch(ttt_file, responsive_ui=True)
pr.step_ui()
robot = Robot(Panda(), PandaGripper())
cam_config = CameraConfig(rgb=True,
depth=False,
mask=False,
render_mode=RenderMode.OPENGL)
obs_config = ObservationConfig()
obs_config.set_all(False)
obs_config.right_shoulder_camera = cam_config
obs_config.left_shoulder_camera = cam_config
obs_config.wrist_camera = cam_config
obs_config.front_camera = cam_config
scene = Scene(pr, robot, obs_config)
loaded_task = LoadedTask(pr, scene, robot)
print(' ,')
print(' /( ___________')
print('| >:===========` Welcome to task builder!')
print(' )(')
print(' ""')
loaded_task.new_task()
while True:
os.system('cls' if os.name == 'nt' else 'clear')
print('\n-----------------\n')
print('The python file will be reloaded when simulation is restarted.')
print('(q) to quit.')
pr.step_ui()
robot = Robot(Panda(), PandaGripper())
cam_config = CameraConfig(rgb=True,
depth=False,
mask=False,
render_mode=RenderMode.OPENGL)
obs_config = ObservationConfig()
obs_config.set_all(False)
obs_config.right_shoulder_camera = cam_config
obs_config.left_shoulder_camera = cam_config
obs_config.overhead_camera = cam_config
obs_config.wrist_camera = cam_config
obs_config.front_camera = cam_config
scene = Scene(pr, robot, obs_config)
loaded_task = LoadedTask(pr, scene, robot)
print(' ,')
print(' /( ___________')
print('| >:===========` Welcome to task builder!')
print(' )(')
print(' ""')
loaded_task.new_task()
while True:
os.system('cls' if os.name == 'nt' else 'clear')
print('\n-----------------\n')
print('The python file will be reloaded when simulation is restarted.')
print('(q) to quit.')
class EnvironmentImpl(Environment):
"""Each environment has a scene."""
@staticmethod
def all_task_names():
return tuple(o[0] for o in getmembers(tasks) if isclass(o[1]))
def __init__(self, task_name: str, obs_config: ObservationConfig = ObservationConfig(task_low_dim_state=True),
action_mode: ActionMode = ActionMode(),
arm_name: str = "Panda", gripper_name: str = "Panda_gripper"):
super(EnvironmentImpl, self).__init__(task_name)
self._arm_name = arm_name
self._gripper_name = gripper_name
self._action_mode = action_mode
self._obs_config = obs_config
# TODO: modify the task/robot/arm/gripper to support early instantiation before v-rep launched
self._task = None
self._pyrep = None
self._robot = None
self._scene = None
self._variation_number = 0
self._reset_called = False
self._prev_ee_velocity = None
self._update_info_dict()
def init(self, display=False):
if self._pyrep is not None:
self.finalize()
with suppress_std_out_and_err():
self._pyrep = PyRep()
# TODO: TTT_FILE should be defined by robot, but now robot depends on launched pyrep
self._pyrep.launch(join(DIR_PATH, TTT_FILE), headless=not display)
self._pyrep.set_simulation_timestep(0.005)
# TODO: Load arm and gripper from name
self._robot = Robot(Panda(), PandaGripper())
self._scene = Scene(self._pyrep, self._robot, self._obs_config)
self._set_arm_control_action()
# Str comparison because sometimes class comparison doesn't work.
if self._task is not None:
self._task.unload()
self._task = self._get_class_by_name(self._task_name, tasks)(self._pyrep, self._robot)
self._scene.load(self._task)
self._pyrep.start()
def finalize(self):
with suppress_std_out_and_err():
self._pyrep.shutdown()
self._pyrep = None
def reset(self, random: bool = True) -> StepDict:
logging.info('Resetting task: %s' % self._task.get_name())
self._scene.reset()
try:
# TODO: let desc be constant
desc = self._scene.init_episode(self._variation_number, max_attempts=MAX_RESET_ATTEMPTS, randomly_place=random)
except (BoundaryError, WaypointError) as e:
raise TaskEnvironmentError(
'Could not place the task %s in the scene. This should not '
'happen, please raise an issues on this task.'
% self._task.get_name()) from e
ctr_loop = self._robot.arm.joints[0].is_control_loop_enabled()
locked = self._robot.arm.joints[0].is_motor_locked_at_zero_velocity()
self._robot.arm.set_control_loop_enabled(False)
self._robot.arm.set_motor_locked_at_zero_velocity(True)
self._reset_called = True
self._robot.arm.set_control_loop_enabled(ctr_loop)
self._robot.arm.set_motor_locked_at_zero_velocity(locked)
# Returns a list o f descriptions and the first observation
return {'s': self._scene.get_observation().get_low_dim_data(), "opt": desc}
def step(self, last_step: StepDict) -> (StepDict, bool):
# returns observation, reward, done, info
if not self._reset_called:
raise RuntimeError("Call 'reset' before calling 'step' on a task.")
assert 'a' in last_step, "Key 'a' for action not in last_step, maybe you passed a wrong dict ?"
# action should contain 1 extra value for gripper open close state
arm_action = np.array(last_step['a'][:-1])
ee_action = last_step['a'][-1]
current_ee = (1.0 if self._robot.gripper.get_open_amount()[0] > 0.9 else 0.0)
if ee_action > 0.0:
ee_action = 1.0
elif ee_action < -0.0:
ee_action = 0.0
if self._action_mode.arm == ArmActionMode.ABS_JOINT_VELOCITY:
self._assert_action_space(arm_action, (len(self._robot.arm.joints),))
self._robot.arm.set_joint_target_velocities(arm_action)
elif self._action_mode.arm == ArmActionMode.DELTA_JOINT_VELOCITY:
self._assert_action_space(arm_action, (len(self._robot.arm.joints),))
cur = np.array(self._robot.arm.get_joint_velocities())
self._robot.arm.set_joint_target_velocities(cur + arm_action)
elif self._action_mode.arm == ArmActionMode.ABS_JOINT_POSITION:
self._assert_action_space(arm_action, (len(self._robot.arm.joints),))
self._robot.arm.set_joint_target_positions(arm_action)
elif self._action_mode.arm == ArmActionMode.DELTA_JOINT_POSITION:
self._assert_action_space(arm_action, (len(self._robot.arm.joints),))
cur = np.array(self._robot.arm.get_joint_positions())
self._robot.arm.set_joint_target_positions(cur + arm_action)
elif self._action_mode.arm == ArmActionMode.ABS_EE_POSE:
self._assert_action_space(arm_action, (7,))
self._ee_action(list(arm_action))
elif self._action_mode.arm == ArmActionMode.DELTA_EE_POSE:
self._assert_action_space(arm_action, (7,))
a_x, a_y, a_z, a_qx, a_qy, a_qz, a_qw = arm_action
x, y, z, qx, qy, qz, qw = self._robot.arm.get_tip().get_pose()
new_rot = Quaternion(a_qw, a_qx, a_qy, a_qz) * Quaternion(qw, qx, qy, qz)
qw, qx, qy, qz = list(new_rot)
new_pose = [a_x + x, a_y + y, a_z + z] + [qx, qy, qz, qw]
self._ee_action(list(new_pose))
elif self._action_mode.arm == ArmActionMode.ABS_EE_VELOCITY:
self._assert_action_space(arm_action, (7,))
pose = self._robot.arm.get_tip().get_pose()
new_pos = np.array(pose) + (arm_action * DT)
self._ee_action(list(new_pos))
elif self._action_mode.arm == ArmActionMode.DELTA_EE_VELOCITY:
self._assert_action_space(arm_action, (7,))
if self._prev_ee_velocity is None:
self._prev_ee_velocity = np.zeros((7,))
self._prev_ee_velocity += arm_action
pose = self._robot.arm.get_tip().get_pose()
pose = np.array(pose)
new_pose = pose + (self._prev_ee_velocity * DT)
self._ee_action(list(new_pose))
elif self._action_mode.arm == ArmActionMode.ABS_JOINT_TORQUE:
self._assert_action_space(arm_action, (len(self._robot.arm.joints),))
self._torque_action(arm_action)
elif self._action_mode.arm == ArmActionMode.DELTA_JOINT_TORQUE:
cur = np.array(self._robot.arm.get_joint_forces())
new_action = cur + arm_action
self._torque_action(new_action)
else:
raise RuntimeError('Unrecognised action mode.')
if current_ee != ee_action:
done = False
while not done:
done = self._robot.gripper.actuate(ee_action, velocity=0.04)
self._pyrep.step()
self._task.step()
if ee_action == 0.0:
# If gripper close action, the check for grasp.
for g_obj in self._task.get_graspable_objects():
self._robot.gripper.grasp(g_obj)
else:
# If gripper opem action, the check for ungrasp.
self._robot.gripper.release()
self._scene.step()
success, terminate = self._task.success()
last_step['r'] = int(success)
next_step = {'s': self._scene.get_observation().get_low_dim_data(), "opt": None}
return last_step, next_step, terminate
def name(self) -> str:
return self._task_name
# ------------- private methods ------------- #
def _update_info_dict(self):
# update info dict
self._info["action mode"] = self._action_mode
self._info["observation mode"] = self._obs_config
# TODO: action dim should related to robot, not action mode, here we fixed it temporally
self._info["action dim"] = (8,)
self._info["action low"] = np.zeros(self._info["action dim"], dtype=np.float32) - 1.
self._info["action high"] = np.zeros(self._info["action dim"], dtype=np.float32) + 1.
self._info["state dim"] = (73,)
self._info["state low"] = np.zeros(self._info["state dim"], dtype=np.float32) - 100.
self._info["state high"] = np.zeros(self._info["state dim"], dtype=np.float32) + 100.
self._info["reward low"] = -np.inf
self._info["reward high"] = np.inf
def _set_arm_control_action(self):
self._robot.arm.set_control_loop_enabled(True)
if self._action_mode.arm in (ArmActionMode.ABS_JOINT_VELOCITY, ArmActionMode.DELTA_JOINT_VELOCITY):
self._robot.arm.set_control_loop_enabled(False)
self._robot.arm.set_motor_locked_at_zero_velocity(True)
elif self._action_mode.arm in (ArmActionMode.ABS_JOINT_POSITION, ArmActionMode.DELTA_JOINT_POSITION,
ArmActionMode.ABS_EE_POSE, ArmActionMode.DELTA_EE_POSE,
ArmActionMode.ABS_EE_VELOCITY, ArmActionMode.DELTA_EE_VELOCITY):
self._robot.arm.set_control_loop_enabled(True)
elif self._action_mode.arm in (ArmActionMode.ABS_JOINT_TORQUE, ArmActionMode.DELTA_JOINT_TORQUE):
self._robot.arm.set_control_loop_enabled(False)
else:
raise RuntimeError('Unrecognised action mode.')
def sample_variation(self) -> int:
self._variation_number = np.random.randint(0, self._task.variation_count())
return self._variation_number
def _assert_action_space(self, action, expected_shape):
if np.shape(action) != expected_shape:
raise RuntimeError(
'Expected the action shape to be: %s, but was shape: %s' % (
str(expected_shape), str(np.shape(action))))
def _torque_action(self, action):
self._robot.arm.set_joint_target_velocities([(TORQUE_MAX_VEL if t < 0 else -TORQUE_MAX_VEL) for t in action])
self._robot.arm.set_joint_forces(np.abs(action))
def _ee_action(self, action):
try:
joint_positions = self._robot.arm.solve_ik(action[:3], quaternion=action[3:])
self._robot.arm.set_joint_target_positions(joint_positions)
except IKError as e:
raise InvalidActionError("Could not find a path.") from e
self._pyrep.step()
@staticmethod
def _get_class_by_name(class_name: str, model: ModuleType) -> TaskClass:
all_class_dict = {}
for o in getmembers(model):
if isclass(o[1]):
all_class_dict[o[0]] = o[1]
if class_name not in all_class_dict:
raise NotImplementedError(f"No class {class_name} found in {model.__name__} !")
return all_class_dict[class_name]