Poster

Controlling Overestimation Bias with Truncated Mixture of Continuous Distributional Quantile Critics

Arsenii Kuznetsov · Pavel Shvechikov · Alexander Grishin · Dmitry Vetrov

Keywords: [ Deep Reinforcement Learning ] [ Reinforcement Learning - Deep RL ]

[ Abstract ] [ Join Zoom
[ Slides
Please do not share or post zoom links

Abstract:

The overestimation bias is one of the major impediments to accurate off-policy learning. This paper investigates a novel way to alleviate the overestimation bias in a continuous control setting. Our method---Truncated Quantile Critics, TQC,---blends three ideas: distributional representation of a critic, truncation of critics prediction, and ensembling of multiple critics. Distributional representation and truncation allow for arbitrary granular overestimation control, while ensembling provides additional score improvements. TQC outperforms the current state of the art on all environments from the continuous control benchmark suite, demonstrating 25% improvement on the most challenging Humanoid environment.

Chat is not available.