📝 SummarXiv

Abstract

Finding the reachable set of a system has a wide range of applications, but is a fundamental challenge in control theory, especially when controls are bounded. Although one can simply integrate the system samples forward in time by applying random admissible control to approximate the reachable set, the samples typically cluster near an attractor (if one is present) -- yielding a poor representation of the reachable set. A better representation can be found by applying controls that specifically lead to a uniform terminal state distribution, however, finding such controls is non-trivial. To find such controls, one must solve an Optimal Transport (OT) problem with uniform measure as the target distribution, which is difficult since the reachable set is not know \emph{a priori}. We can overcome this difficulty by softening the terminal measure constraint via the introduction of a $L_2$-entropy function in the objective and can further reduce this infinite-dimensional regularized OT to a finite-dimensional particle-based optimal control problem by using a nonlocal kernel regularization of the entropy. This leads to a hierarchy of optimization problems whose solutions converge to the original reachability sampling OT problem, as proved by $\Gamma$-convergence. The effectiveness of this entropy-regularized particle-based approach for uniform sampling of reachable set is demonstrated using numerical examples.