SQT -- rough conservative actor critic

Nitsan Soffair, Gilad Katz, Dotan Di Castro, Orly Avner, Shie Mannor

Keywords: Actor Critic, Overestimation Bias

Abstract:

Std $Q$-target is a conservative actor critic ensemble based $Q$-learning algorithm which based on a single key $Q$-formula--$Q$-networks standard deviation, an uncertainty penalty. A minimalistic solution to the problem of overestimation bias. We implement SQT on top of actor critic and test it against the SOTA actor critic algorithms on popular MuJoCo tasks. SQT shows a clear performance advantage over TD3, SAC and TD7 on the tested tasks majority.

Primary Area: reinforcement learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 493