AI Paper Atlas

Research map / Reinforcement learning / Offline RL

Sample Efficient Offline RL via T-Symmetry Enforced Latent State-Stitching

Peng Cheng, Zhihao Wu, Jianxiong Li, Ziteng He, Haoran Xu, Wei Sun, Youfang Lin, Yunxin Liu, Xianyuan Zhan

ICLR 2026 · Reinforcement learning › Offline RL · approach: Simple practical offline RL

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
add latent forward and reverse dynamics to vanilla inverse dynamics module (ablation)TS-IDM · Hopper-medium-expert (10k)normalized score: 17.2 -> 35.5Improved
add ODE property loss (ablation)TS-IDM · Hopper-medium-expert (10k)normalized score: 35.5 -> 61.4Improved
add T-symmetry consistency loss (ablation)TS-IDM · Hopper-medium-expert (10k)normalized score: 61.4 -> 100.9Improved
add T-symmetry consistency loss (ablation)TS-IDM · Halfcheetah-medium-expert (10k)normalized score: 31.2 -> 40.7Improved
add T-symmetry consistency loss (ablation)TS-IDM · Walker2d-medium-expert (10k)normalized score: 58.5 -> 87.4Improved
add latent dynamics modules on Halfcheetah (ablation)TS-IDM · Halfcheetah-medium-expert (10k)normalized score: 29.7 -> 31.3No clear effect
add T-symmetry consistency loss (ablation)TS-IDM · Antmaze-medium-diverse (100k)normalized score: 34.1 -> 47.2Improved
add T-symmetry consistency loss (ablation)TS-IDM · Antmaze-large-play (100k)normalized score: 22.6 -> 47.3Improved
add T-symmetry regularizer in guide-policy optimization (ablation)TELS · 10k datasetsnormalized scoreImproved
replace AE/VAE/contrastive representations with TS-IDM representation (ablation)TELS · 10k datasetsnormalized scoreImproved
use shared weight 1 for all TS-IDM loss terms instead of 0.1 on T-symmetry (ablation)TELS · Hopper-medium-expert (10k)normalized score: 56.4 -> 100.9Improved
use beta 1 instead of 10 (ablation)TELS · Hopper-medium-expert (10k)normalized score: 37.6 -> 100.9Improved

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