Provably Safe Reinforcement Learning via Action Projection Using Reachability Analysis and Polynomial Zonotopes
Top Cited Papers
Open Access
- 13 March 2023
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Open Journal of Control Systems
- Vol. 2, 79-92
Abstract
While reinforcement learning produces very promising results for many applications, its main disadvantage is the lack of safety guarantees, which prevents its use in safety-critical systems. In this work, we address this issue by a safety shield for nonlinear continuous systems that solve reach-avoid tasks. Our safety shield prevents applying potentially unsafe actions from a reinforcement learning agent by projecting the proposed action to the closest safe action. This approach is called action projection and is implemented via mixed-integer optimization. The safety constraints for action projection are obtained by applying parameterized reachability analysis using polynomial zonotopes, which enables to accurately capture the nonlinear effects of the actions on the system. In contrast to other state-of-the-art approaches for action projection, our safety shield can efficiently handle input constraints and dynamic obstacles, eases incorporation of the spatial robot dimensions into the safety constraints, guarantees robust safety despite process noise and measurement errors, and is well suited for high-dimensional systems, as we demonstrate on several challenging benchmark systems.Keywords
All Related Versions
Funding Information
- European Research Council
- Project justITSELF (817629)
- German Research Foundation through the research training group ConVeY (GRK 2428)
- Air Force Office of Scientific Research and the Office of Naval Research (FA9550-19-1-0288, FA9550-21-1-0121, FA9550-23-1-0066, N00014-22-1-2156)
This publication has 52 references indexed in Scilit:
- SPOT: A tool for set-based prediction of traffic participants2017
- CommonRoad: Composable benchmarks for motion planning on roads2017
- Constrained zonotopes: A new tool for set-based estimation and fault detectionAutomatica, 2016
- Safety and Progress for Distributed Cyber-Physical Systems with Unreliable CommunicationACM Transactions on Embedded Computing Systems, 2015
- Computing Large Convex Regions of Obstacle-Free Space Through Semidefinite Programming2015
- A time-dependent Hamilton-Jacobi formulation of reachable sets for continuous dynamic gamesIEEE Transactions on Automatic Control, 2005
- Robust Monte Carlo localization for mobile robotsArtificial Intelligence, 2001
- Disjunctive programming: Properties of the convex hull of feasible pointsDiscrete Applied Mathematics, 1998
- The Simplex architecture for safe online control system upgrades1998
- On the complexity of integer programmingJournal of the ACM, 1981