Learning-Based Shielding for Safe Autonomy under Unknown Dynamics

TL;DR

This paper introduces a data-driven shielding approach for safe autonomous control of unknown systems using deep kernel learning and finite-state abstractions, enabling safety guarantees without known models.

Contribution

It presents a novel methodology combining deep kernel learning with Interval MDPs to provide safety assurances for unknown, continuous-state systems under black-box controllers.

Findings

Guarantees safety for unknown systems with high confidence.

Demonstrates effectiveness on nonlinear and high-dimensional systems.

Provides theoretical proofs of soundness and complexity.

Abstract

Shielding is a common method used to guarantee the safety of a system under a black-box controller, such as a neural network controller from deep reinforcement learning (DRL), with simpler, verified controllers. Existing shielding methods rely on formal verification through Markov Decision Processes (MDPs), assuming either known or finite-state models, which limits their applicability to DRL settings with unknown, continuous-state systems. This paper addresses these limitations by proposing a data-driven shielding methodology that guarantees safety for unknown systems under black-box controllers. The approach leverages Deep Kernel Learning to model the systems' one-step evolution with uncertainty quantification and constructs a finite-state abstraction as an Interval MDP (IMDP). By focusing on safety properties expressed in safe linear temporal logic (safe LTL), we develop an algorithm that computes the maximally permissive set of safe policies on the IMDP, ensuring avoidance of unsafe states. The algorithms soundness and computational complexity are demonstrated through theoretical proofs and experiments on nonlinear systems, including a high-dimensional autonomous spacecraft scenario.