Control DOFs

Position control uses a target per DOF plus proportional and derivative gains.

targets = np.zeros(robot.num_dofs(), dtype=np.float32)

world.set_cmd(
    None,
    obj_id,
    targets,
    mode=ke.sim.ControlMode.POS,
    kp=120.0,
    kd=12.0,
)
world.step(substeps=4)

The first argument selects environments. Use an integer for one environment, None for all environments, or a sequence/tensor for a selected batch.

Available command modes include position, velocity, torque, and the explicit PD path used by advanced GPU control. Start with ControlMode.POS for an MJCF robot with configured joint limits.

Run:

python ./python/examples/mjcf_dof_control.py /path/to/robot.xml

The example exposes every DOF as an ImGui slider and also demonstrates animated targets, reset, collision visualization, and force dragging.

The MJCF articulation guide includes the complete mjcf_dof_control.py source.