Safety through Permissibility: Shield Construction for Fast and Safe Reinforcement Learning
On the map: Deep reinforcement learning methods → Safe and constrained RL
Abstract (arXiv)
Designing Reinforcement Learning (RL) solutions for real-life problems remains a significant challenge. A major area of concern is safety. "Shielding" is a popular technique to enforce safety in RL by turning user-defined safety specifications into safe agent behavior. However, these methods either suffer from extreme learning delays, demand extensive human effort in designing models and safe domains in the problem, or require pre-computation. In this paper, we propose a new permissibility-based framework to deal with safety and shield construction. Permissibility was originally designed for eliminating (non-permissible) actions that will not lead to an optimal solution to improve RL training efficiency. This paper shows that safety can be naturally incorporated into this framework, i.e. extending permissibility to include safety, and thereby we can achieve both safety and improved efficiency. Experimental evaluation using three standard RL applications shows the effectiveness of the approach.
Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in Safe and constrained RL, one of 15 directions in Deep reinforcement learning methods. Builds on 22 of its 37 references that are in the map. First posted on arXiv 29 May 2024.
Builds on
22 of 37 references.
References · 22
- 2015Continuous control with deep reinforcement learningcited by 3,407
- 2015Deep Reinforcement Learning with Double Q-learningcited by 1,745
- 2017Safe Reinforcement Learning via Shieldingcited by 449
- 2019Challenges of Real-World Reinforcement Learningcited by 291
- 2019End-to-End Safe Reinforcement Learning through Barrier Functions for Safety-Critical Continuous Control Taskscited by 289
- 2020A Closer Look at Invalid Action Masking in Policy Gradient Algorithmscited by 171
- 2018Learn What Not to Learn: Action Elimination with Deep Reinforcement Learningcited by 77
- 2021Safe Multi-Agent Reinforcement Learning via Shieldingcited by 61
- 2020Neurosymbolic Reinforcement Learning with Formally Verified Explorationcited by 49
- 2019Robust Model Predictive Shielding for Safe Reinforcement Learning with Stochastic Dynamicscited by 47
- 2020Action Space Shaping in Deep Reinforcement Learningcited by 46
- 2019Learning Robotic Manipulation through Visual Planning and Actingcited by 41
10 more
- 2020What can I do here? A Theory of Affordances in Reinforcement Learningcited by 35
- 2015Safety-Constrained Reinforcement Learning for MDPscited by 33
- 2016End-to-End Deep Reinforcement Learning for Lane Keeping Assistcited by 27
- 2020Tightening Exploration in Upper Confidence Reinforcement Learningcited by 25
- 2020Adaptive Shielding under Uncertaintycited by 18
- 2017Symmetry Learning for Function Approximation in Reinforcement Learningcited by 14
- 2021Shielding Atari Games with Bounded Presciencecited by 14
- 2021Action Redundancy in Reinforcement Learningcited by 9
- 2019Learning Transferable Domain Priors for Safe Exploration in Reinforcement Learningcited by 7
- 2019I'm sorry Dave, I'm afraid I can't do that, Deep Q-learning from forbidden actioncited by 4
Built on by
Citing papers per year: 2024: 1.
1 paper.
Similar, not cited
Close in wording, no citation either way
- 2025Probabilistic Shielding for Safe Reinforcement Learning
- 2024Safe Reinforcement Learning in Black-Box Environments via Adaptive Shielding
- 2024A Dynamic Safety Shield for Safe and Efficient Reinforcement Learning of Navigation Tasks
- 2025CBF-RL: Safety Filtering Reinforcement Learning in Training with Control Barrier Functions
- 2024A Survey of Constraint Formulations in Safe Reinforcement Learning
- 2024Leveraging Approximate Model-based Shielding for Probabilistic Safety Guarantees in Continuous Environments
- 2025Safety Modulation: Enhancing Safety in Reinforcement Learning through Cost-Modulated Rewards
- 2025Provably Optimal Reinforcement Learning under Safety Filtering