Implementation:Haosulab ManiSkill PDJointPosVelController

Knowledge Sources	Haosulab_ManiSkill
Domains	Robotics, Simulation
Last Updated	2026-02-15 08:00 GMT

Overview

Concrete combined PD joint position and velocity controller that simultaneously sets both position and velocity drive targets in ManiSkill.

Description

PDJointPosVelController extends PDJointPosController to support simultaneous position and velocity drive targets. The action space is double the joint dimensions: the first half specifies position targets (absolute or delta, following the same logic as PDJointPosController) and the second half specifies velocity targets that are set directly via set_joint_drive_velocity_targets(). The controller sets both sets_target_qpos = True and sets_target_qvel = True. On reset, the velocity target is initialized to the current joint velocity (or zeros for reset environments in batched simulation).

Usage

Use PDJointPosVelControllerConfig for tasks that benefit from feedforward velocity commands alongside position targets, such as trajectory tracking or dynamic manipulation tasks. The combined position+velocity action provides finer-grained control compared to position-only controllers.

Code Reference

Source Location

Repository: Haosulab_ManiSkill
File: mani_skill/agents/controllers/pd_joint_pos_vel.py
Lines: 1-71

Signature

class PDJointPosVelController(PDJointPosController):
    config: "PDJointPosVelControllerConfig"
    _target_qvel = None
    sets_target_qpos = True
    sets_target_qvel = True

@dataclass
class PDJointPosVelControllerConfig(PDJointPosControllerConfig):
    controller_cls = PDJointPosVelController
    vel_lower: Union[float, Sequence[float]] = -1.0
    vel_upper: Union[float, Sequence[float]] = 1.0

Import

from mani_skill.agents.controllers.pd_joint_pos_vel import (
    PDJointPosVelController,
    PDJointPosVelControllerConfig,
)

I/O Contract

Inputs

Name	Type	Required	Description
action	np.ndarray	Yes	Action of shape (batch, 2 * num_joints). First half is position targets, second half is velocity targets.
vel_lower	float or Sequence[float]	No	Lower bound for velocity targets. Defaults to -1.0.
vel_upper	float or Sequence[float]	No	Upper bound for velocity targets. Defaults to 1.0.
(inherited)			All parameters from `PDJointPosControllerConfig` (lower, upper, stiffness, damping, etc.).

Outputs

Name	Type	Description
_target_qpos	torch.Tensor	Computed position targets set as joint drive position targets.
_target_qvel	torch.Tensor	Velocity targets set as joint drive velocity targets.
action_space	spaces.Box	Box space of dimension 2 * num_joints, concatenating position and velocity bounds.

Usage Examples

Basic Usage

from mani_skill.agents.controllers.pd_joint_pos_vel import PDJointPosVelControllerConfig

# Combined position + velocity control
config = PDJointPosVelControllerConfig(
    joint_names=["joint1", "joint2", "joint3"],
    lower=-3.14,
    upper=3.14,
    stiffness=100,
    damping=10,
    vel_lower=-1.0,
    vel_upper=1.0,
    use_delta=True,
    normalize_action=True,
)
# Action space will be Box of shape (6,): [pos1, pos2, pos3, vel1, vel2, vel3]

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment