Principle:ARISE Initiative Robosuite Joint Velocity Control
| Knowledge Sources | |
|---|---|
| Domains | Robotics, Control Theory, Servo Control |
| Last Updated | 2026-02-15 07:00 GMT |
Overview
Joint velocity control uses PID feedback to compute torques that track desired joint velocity setpoints, converting velocity-level commands into actuator torques with integral anti-windup protection.
Description
Joint velocity control operates at the velocity level of the joint space, accepting desired joint velocities as inputs and computing the torques necessary to track those velocities. This is in contrast to position control (which tracks joint angles) and torque control (which directly commands torques). Velocity control is particularly useful for continuous-motion tasks where specifying a desired rate of motion is more natural than specifying discrete positions.
The control law employs a full PID (proportional-integral-derivative) controller acting on the velocity error e = v_des - v_act for each joint. The proportional term provides the primary corrective torque proportional to the instantaneous velocity error. The integral term accumulates past errors to eliminate steady-state offsets. The derivative term acts on the rate of change of the error signal (smoothed over a ring buffer) to dampen oscillations and improve transient response.
A critical feature is the anti-windup mechanism: the integral term is only accumulated when the controller output is not saturated (i.e., the computed torques are within actuator limits). This prevents the integral state from growing unboundedly when the actuator cannot produce the commanded torque, which would otherwise cause large overshoots when the constraint is released. The gain structure scales the proportional gain by the actuator range to normalize behavior across robots with different torque capacities.
Usage
Apply joint velocity control when the task is best expressed as a desired joint velocity profile, such as maintaining a constant end-effector velocity during wiping or scanning motions, or when a learned policy outputs velocity commands. It is also useful as an inner loop for mobile base control where wheel velocities are the natural control input. Velocity control provides smoother motion than position control for continuous tasks but does not inherently regulate position.
Theoretical Basis
PID Joint Velocity Control Law:
Given:
- Desired joint velocities: v_des (N-dimensional)
- Current joint velocities: v_act (from simulator)
- Proportional gain: K_p (scaled by actuator range)
- Integral gain: K_i = K_p * 0.005
- Derivative gain: K_d = K_p * 0.001
- Ring buffer for derivative smoothing (length 5)
- Gravity compensation: g(q)
1. Compute velocity error:
e = v_des - v_act
2. Compute derivative of error:
de = e - e_prev
de_smooth = average(ring_buffer(de))
3. Update integral (with anti-windup):
if not saturated:
sum_e += e
4. Compute torques:
tau = K_p * e + K_i * sum_e + K_d * de_smooth + g(q)
5. Clip torques and detect saturation:
tau_clipped = clip(tau, tau_min, tau_max)
saturated = (tau_clipped != tau)
Key design decisions:
- PID vs. PD: The integral term is necessary for velocity control because, unlike position control where gravity compensation naturally handles steady-state, velocity tracking requires integral action to overcome friction and modeling errors.
- Anti-windup: Disabling integral accumulation during saturation prevents large transients when the actuator exits its limit.
- Derivative smoothing: Averaging the error derivative over a ring buffer of recent values reduces sensitivity to measurement noise.
- Gain scaling: Proportional gain is automatically scaled by the actuator range, providing consistent behavior across robots with different torque capacities.
- Velocity limits: Optional velocity limits clip the scaled goal velocity before tracking, providing a safety boundary.