Contents

polysim.envs package

Subpackages

Submodules

polysim.envs.abstract_env module

class polysim.envs.abstract_env.AbstractControlledEnv

Bases: ABC

Parent class for simulation environments.

abstract apply_joint_torques(torques)
input:

np.ndarray: Desired torques

abstract get_current_joint_pos_vel()
Returns:

Joint positions np.ndarray: Joint velocities

Return type:

np.ndarray

abstract get_current_joint_torques()
Returns:

Torques received from apply_joint_torques np.ndarray: Torques sent to robot (e.g. after clipping) np.ndarray: Torques generated by the actuators (e.g. after grav comp) np.ndarray: Torques exerted onto the robot

Return type:

np.ndarray

abstract get_num_dofs()

Get the number of degrees of freedom for controlling the simulation.

Returns:

Number of control input dimensions

Return type:

int

abstract reset()

Reset the environment.

abstract set_robot_state(robot_state)
input:

RobotState: Desired state of robot

polysim.envs.bullet_manipulator module

class polysim.envs.bullet_manipulator.BulletManipulatorEnv(robot_model_cfg: DictConfig, gui: bool, use_grav_comp: bool = True, gravity: float = 9.81, extract_config_from_rdf=False)

Bases: AbstractControlledEnv

A manipulator environment using PyBullet.

Parameters:

  • robot_model_cfg – A Hydra configuration file containing information needed for the robot model, e.g. URDF. For an example, see polymetis/conf/robot_model/franka_panda.yaml

  • gui – Whether to initialize the PyBullet simulation in GUI mode.

  • use_grav_comp – If True, adds gravity compensation torques to the input torques.

apply_joint_torques(torque: ndarray)

Applies a NumPy array of torques and returns the final applied torque (after gravity compensation, if used).

compute_forward_kinematics(joint_pos: Optional[List[float]] = None)

Computes forward kinematics.

Warning

Uses PyBullet forward kinematics by resetting to the given joint position (or, if not given, the rest position). Therefore, drastically modifies simulation state!

Parameters:

joint_pos – Joint positions for which to compute forward kinematics.

Returns:

3-dimensional end-effector position

np.ndarray: 4-dimensional end-effector orientation as quaternion

Return type:

np.ndarray

compute_inverse_dynamics(joint_pos: ndarray, joint_vel: ndarray, joint_acc: ndarray)

Computes inverse dynamics by returning the torques necessary to get the desired accelerations at the given joint position and velocity.

compute_inverse_kinematics(target_position: List[float], target_orientation: Optional[List[float]] = None)

Computes inverse kinematics.

Uses PyBullet to compute inverse kinematics.

Parameters:

  • target_position – Desired end-effector position.

  • target_orientation – Desired end-effector orientation as quaternion.

Returns:

Joint position satisfying the given target end-effector position/orientation.

Return type:

np.ndarray

get_current_joint_pos()

Returns current joint position as a NumPy array.

get_current_joint_pos_vel()

Returns (current joint position, current joint velocity) as a tuple of NumPy arrays

get_current_joint_torques()

Returns torques: [inputted, clipped, added with gravity compensation, and measured externally]

get_current_joint_vel()

Returns current joint velocity as a NumPy array.

get_num_dofs()

Return number of degrees of freedom for control

static load_robot_description_from_sdf(abs_urdf_path: str, sim: BulletClient)

Loads a SDF file into the simulation.

static load_robot_description_from_urdf(abs_urdf_path: str, sim: BulletClient)

Loads a URDF file into the simulation.

reset(joint_pos: Optional[List[float]] = None, joint_vel: Optional[List[float]] = None)

Resets simulation to the given pose, or if not given, the default rest pose

set_robot_state(robot_state)
input:

RobotState: Desired state of robot

Module contents

polysim package torchcontrol package