AI Paper Atlas

Research map / Reinforcement learning / Offline RL

Distributional Soft Actor-Critic With Three Refinements

Jingliang Duan, Wenxuan Wang, Liming Xiao, Jiaxin Gao, Shengbo Eben Li, Chang Liu, Ya-Qin Zhang, Bo Cheng, Keqiang Li

TPAMI 2025 · Reinforcement learning › Offline RL · approach: Offline RL sample complexity theory · 122 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
apply the three refinements (DSAC-T) to DSACv1DSAC · Humanoid-v3average final return: 9074 -> 10829Improved
apply the three refinements (DSAC-T) to DSACv1DSAC · BipedalWalker-hardcore-v3average final return: -31 -> 77Improved
use DSAC-T instead of SACSAC · Humanoid-v3average final return: 9335 -> 10829Improved
use DSAC-T instead of SACSAC · Swimmer-v3average final return: 140 -> 138No clear effect
add twin value distribution learning (lower-Q distribution) to DSACDSAC · Humanoid-v3Q-value estimation bias: 207.25 -> -42.29Improved
add twin value distribution learning (lower-Q distribution) to DSACDSAC · Hopper-v3Q-value estimation bias: 2171.96 -> -7.00Improved
use DSAC-T instead of RAC-SACRAC-SAC · Humanoid-v3Q-value estimation bias: 97.78 -> -42.29Improved
use DSAC-T instead of RAC-SACRAC-SAC · Ant-v3Q-value estimation bias: -18.82 -> -10.55Improved
add expected value substituting to DSAC-T (ablation)DSAC-T · Humanoid-v3learning stability and performanceImproved
add twin value distribution learning to DSAC-T (ablation)DSAC-T · Humanoid-v3learning stability and performanceImproved
add variance-based critic gradient adjustment to DSAC-T (ablation)DSAC-T · HalfCheetah-v3 (reward scales 0.01 and 0.1)reward-scale robustnessImproved
use DSAC-T instead of DSACv1 in compute costDSAC · Humanoid-v3seconds per 1000 iterations: 29.00 -> 35.51Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy