Certainty Equivalent Perception-Based Control
This work addresses safety certification for control systems using learned sensors, which is crucial for robotics and autonomous vehicles, but it appears incremental as it builds on existing kernel regression methods.
The paper tackles the problem of certifying performance and safety in feedback control systems when sensors rely on supervised learning, by providing a uniform error bound for nonparametric kernel regression under dense sampling and showing finite-time convergence rates for waypoint tracking. It demonstrates results in simulation with UAV and autonomous driving examples, though no concrete numbers are specified.
In order to certify performance and safety, feedback control requires precise characterization of sensor errors. In this paper, we provide guarantees on such feedback systems when sensors are characterized by solving a supervised learning problem. We show a uniform error bound on nonparametric kernel regression under a dynamically-achievable dense sampling scheme. This allows for a finite-time convergence rate on the sub-optimality of using the regressor in closed-loop for waypoint tracking. We demonstrate our results in simulation with simplified unmanned aerial vehicle and autonomous driving examples.