Finite sample learning of moving targets
This work addresses the challenge of learning dynamic systems in real-time for applications like autonomous driving, though it appears incremental as it builds on existing techniques for constant targets.
The paper tackles the problem of learning a moving target from finite samples, extending randomized techniques from constant targets to changing ones and deriving a novel bound on the number of samples needed for a probably approximately correct (PAC) estimate. It provides a constructive method using mixed integer linear programming for convex polytope targets and demonstrates it on autonomous emergency braking.
We consider a moving target that we seek to learn from samples. Our results extend randomized techniques developed in control and optimization for a constant target to the case where the target is changing. We derive a novel bound on the number of samples that are required to construct a probably approximately correct (PAC) estimate of the target. Furthermore, when the moving target is a convex polytope, we provide a constructive method of generating the PAC estimate using a mixed integer linear program (MILP). The proposed method is demonstrated on an application to autonomous emergency braking.