dm_robotics.panda.arm module

MoMa robot arm model of the Panda along with specialized effectors and sensors.

class dm_robotics.panda.arm.ArmEffector(robot_params: RobotParams, arm: RobotArm)

Bases: ArmEffector

Robot arm effector for the Panda MoMa model.

Takes dm_robotics.panda.parameters.RobotParams and changes the joint stiffness and damping of the robot arm. Otherwise behaves like dm_robotics.moma.effectors.arm_effector.ArmEffector.

__init__(robot_params: RobotParams, arm: RobotArm)

Constructor.

Parameters:

• robot_params – Dataclass containing robot parameters.

• arm – The MoMa arm to control.

after_compile(mjcf_model: RootElement, physics: Physics) → None

Method called after the MJCF model has been compiled and finalized.

Parameters:

• mjcf_model – The root element of the scene MJCF model.

• physics – Compiled physics.

set_control(physics: Physics, command: ndarray) → None

class dm_robotics.panda.arm.Cartesian6dVelocityEffector(robot_params: RobotParams, arm: RobotArm, gripper: RobotHand, joint_velocity_effector: Effector, tcp_sensor: RobotTCPSensor, control_timestep: float = 0.1)

Bases: Cartesian6dVelocityEffector

Panda Version of the MoMa Cartesian6dVelocityEffector.

__init__(robot_params: RobotParams, arm: RobotArm, gripper: RobotHand, joint_velocity_effector: Effector, tcp_sensor: RobotTCPSensor, control_timestep: float = 0.1)

Initializes a QP-based 6D Cartesian velocity effector.

Parameters:

• robot_name – name of the robot the Cartesian effector controls.

• joint_velocity_effector – Effector on the joint velocities being controlled to achieve the target Cartesian velocity. This class takes ownership of this effector, i.e. it will call initialize_episode automatically.

• model_params – parameters that describe the object being controlled.

• control_params – parameters that describe how the element should be controlled.

• collision_params – parameters that describe the active collision avoidance behaviour, if any.

• log_nullspace_failure_warnings – if true, a warning will be logged if the internal LSQP solver is unable to solve the nullspace optimization problem (second hierarchy). Ignored if nullspace control is disabled.

• use_adaptive_qp_step_size – if true, the internal LSQP solver will use an adaptive step size when solving the resultant QP problem. Note that setting this to true can greatly speed up the computation time, but the solution will no longer be numerically deterministic.

• compile_mjcf – if true, it will compile an MjModel from the MJCF during the after_compile stage. If false, it will re-use the MjModel in the mjcf.Physics object passed during the after_compile stage. Should only be set to false if the MjModel within the mjcf.Physics is guaranteed to be finalized and constant before and after after_compile is called.

set_control(physics: Physics, command: ndarray) → None

Sets a 6 DoF Cartesian velocity command at the current timestep.

Parameters:

• physics – mjcf.Physics object with the updated environment state at the current timestep.

• command – array of size 6 describing the desired 6 DoF Cartesian target [(lin_vel), (ang_vel)].

class dm_robotics.panda.arm.ControlObservations(value)

Bases: Enum

Observations exposed by this sensor in control frame.

POS = '{}_pos_control'

POSE = '{}_pose_control'

QUAT = '{}_quat_control'

RMAT = '{}_rmat_control'

VEL = '{}_vel_control'

get_obs_key(name: str) → str

Returns the key to the observation in the observables dict.

class dm_robotics.panda.arm.ExternalWrenchObserver(robot_params: RobotParams, arm: Panda, arm_sensor: RobotArmSensor)

Bases: Sensor

Estimates external wrench based on torque sensor signal.

__init__(robot_params: RobotParams, arm: Panda, arm_sensor: RobotArmSensor) → None

after_compile(mjcf_model: RootElement, physics: Physics) → None

Method called after the MJCF model has been compiled and finalized.

Parameters:

• mjcf_model – The root element of the scene MJCF model.

• physics – Compiled physics.

get_obs_key(obs: Enum) → str

Returns the key to an observable provided by this sensor.

initialize_episode(physics: Physics, random_state: RandomState) → None

Called on a new episode, after the environment has been reset.

This is called before the agent has got a timestep in the episode that is about to start. Sensors can reset any state they may have.

Parameters:

• physics – The MuJoCo physics the environment uses.

• random_state – A PRNG seed.

property name: str

property observables: Dict[str, Observable]

Get the observables for this Sensor.

This will be called after initialize_for_task.

It’s expected that the keys in this dict are values from self.get_obs_key(SOME_ENUM_VALUE), the values are Observables. See the class docstring for more information about Observable.

subclassing dm_control.composer.observation.observable.Generic is a simple way to create an Observable. Observables have many properties that are used by composer to alter how the values it produces are processed.

See the code dm_control.composer.observation for more information.

class dm_robotics.panda.arm.Panda(*args, **kwargs)

Bases: RobotArm

A class representing a Panda robot arm.

Parameters:

• name – The name of this robot. Used as a prefix in the MJCF name attributes.

• actuation – Instance of dm_robotics.panda.arm_constants.Actuation specifying which actuation mode to use.

• use_rotated_gripper – If True, mounts the gripper in a rotated position to match the real placement of the gripper on the physical Panda.

property actuation: Actuation

Selected actuation mode.

property actuators: List[_ElementImpl]

List of actuator elements belonging to the arm.

after_substep(physics: Physics, random_state: RandomState) → None

A callback which is executed after a simulation step.

property attachment_site

Override wrist site for attachment, but NOT the one for observations.

property base_site: _ElementImpl

Get the MuJoCo site of the base.

Returns:

MuJoCo site

initialize_episode(physics: Physics, random_state: RandomState)

Function called at the beginning of every episode.

property joint_torque_sensors: List[_ElementImpl]

Get MuJoCo sensor of the joint torques.

property joints: List[_ElementImpl]

List of joint elements belonging to the arm.

property mjcf_model: RootElement

Returns the mjcf.RootElement object corresponding to this robot arm.

property name: str

Name of the robot arm.

set_joint_angles(physics: Physics, joint_angles: ndarray) → None

Sets the joints of the robot to a given configuration.

This function allows to change the joint configuration of the Panda arm and sets the controller to prevent the impedance controller from moving back to the previous configuration.

Parameters:

• physics – A mujoco.Physics instance.

• joint_angles – The desired joints configuration for the robot arm.

property wrist_site: _ElementImpl

Get the MuJoCo site of the wrist.

Returns:

MuJoCo site

class dm_robotics.panda.arm.RobotArmSensor(robot_params: RobotParams, arm: RobotArm)

Bases: RobotArmSensor

Behaves like dm_robotics.moma.sensors.robot_arm_sensor.RobotArmSensor.

Except that the joint torque signal does not include passive forces. This is done so as to model the external torque signal provided by the Panda robot.

__init__(robot_params: RobotParams, arm: RobotArm)

class dm_robotics.panda.arm.RobotTCPSensor(gripper: RobotHand | RobotHand, robot_params: RobotParams)

Bases: SiteSensor

Version of dm_robotics.moma.sensors.site_sensor.SiteSensor that takes tool center point pose measurements of a gripper and accepts parameters.RobotParams to optionally change the reference frame. Otherwise behaves like a SiteSensor.

__init__(gripper: RobotHand | RobotHand, robot_params: RobotParams)

Constructor.

Parameters:

gripper – This gripper’s TCP site is used for the measurements. robot_params: Set the control_frame field to a site to use as reference frame. Falls back to world frame if None.

dm_robotics.panda.arm.build_robot(robot_params: RobotParams, control_timestep: float = 0.1) → Robot

Builds a MoMa robot model of the Panda.