ISAACS: Iterative Soft Adversarial Actor-Critic for Safety

TL;DR

This paper presents ISAACS, a scalable method combining adversarial reinforcement learning and safety analysis to synthesize robust safety-preserving controllers for nonlinear robotic systems operating under uncertain conditions.

Contribution

It introduces a novel iterative soft actor-critic framework that co-trains a safety-seeking policy with an adversarial disturbance agent to ensure robust safety guarantees.

Findings

The safety shield effectively prevents safety violations in simulation.

The learned policy performs comparably to optimal solutions.

Empirical validation confirms robustness against worst-case model errors.

Abstract

The deployment of robots in uncontrolled environments requires them to operate robustly under previously unseen scenarios, like irregular terrain and wind conditions. Unfortunately, while rigorous safety frameworks from robust optimal control theory scale poorly to high-dimensional nonlinear dynamics, control policies computed by more tractable "deep" methods lack guarantees and tend to exhibit little robustness to uncertain operating conditions. This work introduces a novel approach enabling scalable synthesis of robust safety-preserving controllers for robotic systems with general nonlinear dynamics subject to bounded modeling error by combining game-theoretic safety analysis with adversarial reinforcement learning in simulation. Following a soft actor-critic scheme, a safety-seeking fallback policy is co-trained with an adversarial "disturbance" agent that aims to invoke the worst-case realization of model error and training-to-deployment discrepancy allowed by the designer's uncertainty. While the learned control policy does not intrinsically guarantee safety, it is used to construct a real-time safety filter (or shield) with robust safety guarantees based on forward reachability rollouts. This shield can be used in conjunction with a safety-agnostic control policy, precluding any task-driven actions that could result in loss of safety. We evaluate our learning-based safety approach in a 5D race car simulator, compare the learned safety policy to the numerically obtained optimal solution, and empirically validate the robust safety guarantee of our proposed safety shield against worst-case model discrepancy.