Addressing Function Approximation Error in Actor-Critic Methods

Code

• github.com/sfujim/TD3
  OfficialIn paperpytorch★ 0
• github.com/DLR-RM/stable-baselines3
  pytorch★ 12,962
• github.com/facebookresearch/ReAgent
  pytorch★ 3,690
• github.com/toni-sm/skrl
  jax★ 1,014
• github.com/Rafael1s/Deep-Reinforcement-Learning-Udacity
  pytorch★ 992
• github.com/tensorlayer/RLzoo
  tf★ 644
• github.com/intelligent-environments-lab/CityLearn
  tf★ 594
• github.com/araffin/sbx
  jax★ 572
• github.com/andrejorsula/drl_grasping
  pytorch★ 506
• github.com/arrival-ltd/catalyst-rl-tutorial
  pytorch★ 202

Abstract

In value-based reinforcement learning methods such as deep Q-learning, function approximation errors are known to lead to overestimated value estimates and suboptimal policies. We show that this problem persists in an actor-critic setting and propose novel mechanisms to minimize its effects on both the actor and the critic. Our algorithm builds on Double Q-learning, by taking the minimum value between a pair of critics to limit overestimation. We draw the connection between target networks and overestimation bias, and suggest delaying policy updates to reduce per-update error and further improve performance. We evaluate our method on the suite of OpenAI gym tasks, outperforming the state of the art in every environment tested.

Tasks

Benchmark Results

Reproductions