T-PFC — Trajectory-Optimized Perturbation Feedback Control
Published in IEEE Robotics and Automation Letters and presented at IEEE/RSJ IROS 2019, Macau, China.
Overview
Robot motion controllers typically handle trajectory generation and disturbance rejection as separate concerns. T-PFC unifies them: it jointly optimizes the nominal reference trajectory and the perturbation feedback gains, so the planned path is inherently shaped around the system’s ability to handle uncertainty.
The key insight is that a trajectory that looks optimal in the nominal (deterministic) sense may be brittle under real-world noise, while a slightly suboptimal nominal path may enable a much more robust closed-loop response. T-PFC captures this trade-off explicitly in the optimization objective.
Method
- Frames the problem as a stochastic optimal control problem over a finite horizon.
- Uses a perturbation feedback (LQR-style) controller layered on top of a nominal trajectory.
- Jointly optimizes both the nominal trajectory and the feedback gains using iterative LQR (iLQR) with covariance propagation.
Results
- Validated on quadrotor trajectory tracking and robot arm motion tasks.
- Outperforms standard iLQR and MPC baselines in the presence of process noise and model mismatch.