Computed-Torque Trajectory Tracking

Key Insight

Computed-torque control uses a model of the arm's own physics — the manipulator equation — as a feedforward term that cancels its nonlinear dynamics in advance, so the leftover error behaves like a simple, decoupled linear system that a light PID correction can clean up. Tracking a sinusoidal joint trajectory on a 6-DoF arm and comparing PID-only against feedforward-plus-PID makes the lesson concrete: the model-based prediction does the heavy lifting, and feedback only mops up what the model got slightly wrong. This is why a good enough dynamics model, not an ever-higher feedback gain, is what separates a wobbly tracker from a crisp one.