mlsanityData to 14 Oct 2026

ROSARL: Reward-Only Safe Reinforcement Learning

arXiv 2306.00035 ↗First posted 31 May 2023cs.LGcited by 5

Not on the map. Why?

Abstract (arXiv)

An important problem in reinforcement learning is designing agents that learn to solve tasks safely in an environment. A common solution is for a human expert to define either a penalty in the reward function or a cost to be minimised when reaching unsafe states. However, this is non-trivial, since too small a penalty may lead to agents that reach unsafe states, while too large a penalty increases the time to convergence. Additionally, the difficulty in designing reward or cost functions can increase with the complexity of the problem. Hence, for a given environment with a given set of unsafe states, we are interested in finding the upper bound of rewards at unsafe states whose optimal policies minimise the probability of reaching those unsafe states, irrespective of task rewards. We refer to this exact upper bound as the "Minmax penalty", and show that it can be obtained by taking into account both the controllability and diameter of an environment. We provide a simple practical model-free algorithm for an agent to learn this Minmax penalty while learning the task policy, and demonstrate that using it leads to agents that learn safe policies in high-dimensional continuous control environments.

Key facts. Cited by 5 papers in the map (data through 14 Oct 2026). Builds on 16 of its 30 references that are in the map. First posted on arXiv 31 May 2023.

Builds on

16 of 30 references.

References · 16

  1. 2015Trust Region Policy Optimizationcited by 2,580
  2. 2016Concrete Problems in AI Safetycited by 1,642
  3. 2017Constrained Policy Optimizationcited by 983
  4. 2018QT-Opt: Scalable Deep Reinforcement Learning for Vision-Based Robotic Manipulationcited by 743
  5. 2020Deep Reinforcement Learning for Autonomous Driving: A Surveycited by 721
  6. 2017Safe Reinforcement Learning via Shieldingcited by 449
  7. 2020Responsive Safety in Reinforcement Learning by PID Lagrangian Methodscited by 316
  8. 2018A Lyapunov-based Approach to Safe Reinforcement Learningcited by 297
  9. 2018Safe Exploration in Continuous Action Spacescited by 247
  10. 2020Projection-Based Constrained Policy Optimizationcited by 203
  11. 2017Self-supervised Deep Reinforcement Learning with Generalized Computation Graphs for Robot Navigationcited by 77
  12. 2019A Survey of Deep Reinforcement Learning in Video Gamescited by 73
4 more
  1. 2022Saute RL: Almost Surely Safe Reinforcement Learning Using State Augmentationcited by 69
  2. 2021Safe Reinforcement Learning Using Advantage-Based Interventioncited by 47
  3. 2020Learning with Safety Constraints: Sample Complexity of Reinforcement Learning for Constrained MDPscited by 32
  4. 2022Reinforcement Learning with a Terminatorcited by 4

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2023: 1; 2024: 1; 2025: 1; 2026: 2.

5 papers.

  1. 2023SafeDreamer: Safe Reinforcement Learning with World Modelscited by 54
  2. 2025Safety-Aware Reinforcement Learning for Control via Risk-Sensitive Action-Value Iteration and Quantile Regressioncited by 2
  3. 2024Myopically Verifiable Probabilistic Certificates for Safe Control and Learningcited by 1
  4. 2026Solving Parameter-Robust Avoid Problems with Unknown Feasibility using Reinforcement Learningcited by 0
  5. 2026Safe Greenhouse Climate Control Using Lagrangian-Constrained PPO with Kolmogorov-Arnold Networkscited by 0