Safety from in-the-loop reachability for cyber-physical systems
C. Llanes, M. Abate, S. Coogan
(extended abstract), Proceedings of the Workshop on Computation-Aware Algorithmic Design for Cyber-Physical Systems, 2021
Abstract
We demonstrate a methodology for achieving safe autonomy that relies on computing reachable sets at runtime. Given a system subject to disturbances controlled by an unverified and potentially faulty controller, this methodology computes at each time the reachable set of the system under a backup control law to ensure the system is within reach of a known a priori safe region. Control barrier functions are then used in conjunction with the reachable set to adjust potentially unsafe control actions that would otherwise move the system beyond reach of this safe set. This approach faces several computational challenges: reachable sets for the dynamics must be computed at runtime; sensitivity of the reachable set to initial conditions is required for the control barrier optimization formulation; and the presence of disturbances introduces a large number of constraints in the resulting optimization. The proposed methodology leverages the theory of mixed monotone systems to address these challenges, and the main contribution of this paper is an application of this methodology to a ten dimensional dual planar multirotor system that is implemented on embedded hardware with a controller update rate up to 100Hz.