← Robotics Test Bench

Experiment #01 · Resolved

Feedback fundamentals

One joint, one motor, and a small comparison between position-only and position-plus-velocity feedback.

01 · Question

What does closing the loop change?

How should a one-joint pendulum behave when its controller knows only position error versus both position error and velocity?

02 · Prediction

Position correction should need damping.

P-only control should push on position error and tend to overshoot or oscillate. Adding derivative feedback should damp motion and settle more cleanly.

03 · Experiment

Keep the mechanism small.

A MuJoCo pendulum used one rigid link, one hinge, gravity, and one motor. The same script exposed constant, P, and PD control through direct state and control values: qpos, qvel, and ctrl.

04 · What happened

The prediction matched the run.

P-only behavior overshot or oscillated more. PD behavior was visibly more damped and settled more cleanly.

05 · Model update

Position error and velocity answer different questions.

P answers “how far from target?” D adds “how fast am I moving?” That was enough evidence to move on rather than polish this controller.

Source

Canonical evidence