Leveraging Approximate Model-based Shielding for Probabilistic Safety Guarantees in Continuous Environments

2024/02/01 by Alexander W. Goodall, Goodall, Alexander W., Francesco Belardinelli +1 · 3 citations
Decision Sciences · Engineering · #Artificial Intelligence (cs.AI) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Nuclear and radioactivity studies #Probabilistic and Robust Engineering Design #Risk and Safety Analysis

paper · pdf · doi:10.48550/arxiv.2402.00816

openalex publication_date 2024/02/01 · openalex created_date 2024/02/03 · openalex updated_date 2026/07/28

Abstract

Shielding is a popular technique for achieving safe reinforcement learning (RL). However, classical shielding approaches come with quite restrictive assumptions making them difficult to deploy in complex environments, particularly those with continuous state or action spaces. In this paper we extend the more versatile approximate model-based shielding (AMBS) framework to the continuous setting. In particular we use Safety Gym as our test-bed, allowing for a more direct comparison of AMBS with popular constrained RL algorithms. We also provide strong probabilistic safety guarantees for the continuous setting. In addition, we propose two novel penalty techniques that directly modify the policy gradient, which empirically provide more stable convergence in our experiments.

Cited by

• ProSh: Probabilistic Shielding for Model-free Reinforcement Learning
• Constraint-Aware Reinforcement Learning via Adaptive Action Scaling
• Conformal Safety Shielding for Imperfect-Perception Agents

Related

• Probabilistic Shielding for Safe Reinforcement Learning
• ProSh: Probabilistic Shielding for Model-free Reinforcement Learning
• Adaptive Shielding for Safe Reinforcement Learning under Hidden-Parameter Dynamics Shifts
• Safe Reinforcement Learning via Probabilistic Logic Shields
• Safe Reinforcement Learning via Shielding under Partial Observability