AI Paper Atlas

Research map / Reinforcement learning / Offline RL

Peak-Return Greedy Slicing: Subtrajectory Selection for Transformer-based Offline RL

Zhiwei Xu, Miduo Cui, Dapeng Li, Zhihao Liu, Haifeng Zhang, Hangyu Mao, Guoliang Fan, Bin Zhang

ICLR 2026 · Reinforcement learning › Offline RL · approach: Offline RL sample complexity theory

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 PRGS subtrajectory slicing to Decision TransformerDT · D4RL Maze2D (average)normalized return: 40.9 -> 100.1Improved
add PRGS to Decision TransformerDT · D4RL Gym (average)normalized return: 75.3 -> 82.9Improved
add PRGS to PDiTPDiT · D4RL AntMaze (average)normalized return: 38.7 -> 37.1Hurt
add PRGS to DTDT · D4RL Adroit door tasksnormalized return: door-human 22.7 -> 7.8Mixed
add PRGS to BCTransformer BC · AuctionNet (average)score: 269.9 -> 279.0Improved
add PRGS to PDiTPDiT · BabyAI (average)success rate: 63.0 -> 65.9Improved
add PRGS on narrow-coverage random and expert datasetsPDiT · Gym random/expert (average)normalized return: 54.0 -> 54.2No clear effect
add adaptive history truncation at evaluation (ablation)PRGS · Maze2Dnormalized returnImproved
use very small particle number n in top-n optimistic estimate (ablation)PRGS · Maze2Dnormalized returnHurt
use very large n (conservative value estimate) (ablation)PRGS · Maze2Dnormalized returnHurt
train on top 10% or 20% of trajectories instead of PRGS slicing (ablation)BC, DT, PDiT · Maze2Dnormalized returnNo clear effect
use optimistic MMD estimator rather than w/o Optimistic or Q-loss (ablation)PRGS · hopper-medium and hopper-medium-replaynormalized returnImproved

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