Trustworthy Reinforcement Learning Against Intrinsic Vulnerabilities: Robustness, Safety, and Generalizability
It synthesizes existing research to promote trustworthiness in reinforcement learning for real-world applications, but it is incremental as a survey rather than introducing new methods.
This survey paper tackles the problem of making reinforcement learning trustworthy by addressing its intrinsic vulnerabilities in robustness, safety, and generalizability, providing a unified framework that categorizes methodologies and benchmarks without presenting new experimental results.
A trustworthy reinforcement learning algorithm should be competent in solving challenging real-world problems, including {robustly} handling uncertainties, satisfying {safety} constraints to avoid catastrophic failures, and {generalizing} to unseen scenarios during deployments. This study aims to overview these main perspectives of trustworthy reinforcement learning considering its intrinsic vulnerabilities on robustness, safety, and generalizability. In particular, we give rigorous formulations, categorize corresponding methodologies, and discuss benchmarks for each perspective. Moreover, we provide an outlook section to spur promising future directions with a brief discussion on extrinsic vulnerabilities considering human feedback. We hope this survey could bring together separate threads of studies together in a unified framework and promote the trustworthiness of reinforcement learning.