Optimal Safe Policy

Optimal safe policy research focuses on developing algorithms that learn policies guaranteeing safety while maximizing performance, particularly in high-risk domains. Current approaches leverage techniques like constrained reinforcement learning, model predictive control with chance constraints, and neurosymbolic methods incorporating weakest preconditions or safety monitors to ensure safety during both training and deployment. This field is crucial for enabling the safe application of reinforcement learning in real-world settings such as robotics, autonomous driving, and healthcare, where safety is paramount. The development of provably safe and efficient algorithms is driving progress towards reliable and trustworthy AI systems.

Papers

August 21, 2024

CSPI-MT: Calibrated Safe Policy Improvement with Multiple Testing for Threshold Policies
Brian M Cho, Ana-Roxana Pop, Kyra Gan, Sam Corbett-Davies, Israel Nir, Ariel Evnine, Nathan Kallus
Adversarial Scenario Policy Improvement Multiple Hypothesis Testing Optimal Safe Policy

August 25, 2023

Learn With Imagination: Safe Set Guided State-wise Constrained Policy Optimization
Feihan Li, Yifan Sun, Weiye Zhao, Rui Chen, Tianhao Wei, Changliu Liu
Deep Reinforcement Learning Safe Exploration State Constraint Safe Policy Optimal Safe Policy

October 26, 2022

Provable Safe Reinforcement Learning with Binary Feedback
Andrew Bennett, Dipendra Misra, Nathan Kallus
Reinforcement Learning Safe Reinforcement Learning Confidence Oracle Binary Feedback Optimal Safe Policy

September 28, 2022

Guiding Safe Exploration with Weakest Preconditions
Greg Anderson, Swarat Chaudhuri, Isil Dillig
Reinforcement Learning Safety Constraint Safe Exploration Neurosymbolic Approach Optimal Safe Policy

January 20, 2022

Sim-to-Lab-to-Real: Safe Reinforcement Learning with Shielding and Generalization Guarantees
Kai-Chieh Hsu, Allen Z. Ren, Duy Phuong Nguyen, Anirudha Majumdar, Jaime F. Fisac
Reinforcement Learning Autonomous System Safe Reinforcement Learning Based Policy Generalization Guarantee Bellman Equation Novel Shielding Mechanism Optimal Safe Policy

December 27, 2021

Safe Reinforcement Learning with Chance-constrained Model Predictive Control
Samuel Pfrommer, Tanmay Gautam, Alec Zhou, Somayeh Sojoudi
Predictive Control Policy Gradient Safe Reinforcement Learning Real World Reinforcement Learning Optimal Safe Policy

November 15, 2021

Joint Synthesis of Safety Certificate and Safe Control Policy using Constrained Reinforcement Learning
Haitong Ma, Changliu Liu, Shengbo Eben Li, Sifa Zheng, Jianyu Chen
Reinforcement Learning Constrained Reinforcement Learning Safe Control Joint Generative Safety Certificate Optimal Safe Policy

Optimal Safe Policy

Papers

CSPI-MT: Calibrated Safe Policy Improvement with Multiple Testing for Threshold Policies

Learn With Imagination: Safe Set Guided State-wise Constrained Policy Optimization

Provable Safe Reinforcement Learning with Binary Feedback

Guiding Safe Exploration with Weakest Preconditions

Sim-to-Lab-to-Real: Safe Reinforcement Learning with Shielding and Generalization Guarantees

Safe Reinforcement Learning with Chance-constrained Model Predictive Control

Joint Synthesis of Safety Certificate and Safe Control Policy using Constrained Reinforcement Learning