AI Paper Atlas

Research map / Reinforcement learning / Offline RL

When does return-conditioned supervised learning work for offline reinforcement learning?

David Brandfonbrener, Alberto Bietti, Jacob Buckman, Romain Laroche, Joan Bruna

NeurIPS 2022 · Reinforcement learning › Offline RL · approach: Offline-to-online fine-tuning

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
Use dense reward instead of sparse or ring-of-fire rewardRCSL (RvS, DT) · Point-mass navigationreturnImproved
Use ring-of-fire reward, giving very low probability of high-return trajectoriesRCSL (RvS, DT) · Point-mass navigationreturnHurt
Use sparse reward instead of dense rewardRCSL (RvS, DT) · Point-mass navigationreturnHurt
Replace RCSL with dynamic programming (IQL) under dense rewardPoint-mass dense reward · Point-mass navigationreturnImproved
Use TD3+BC instead of IQL on point-mass data with partial state coverageDP methods · Point-mass navigationreturnHurt
Use DP methods (TD3+BC, IQL) instead of RCSL or BC on antmaze with stochastic behaviorD4RL antmaze-umaze and antmaze-medium-play · D4RL antmazenormalized scoreImproved
Use RCSL instead of BC on high-quality dense-reward data with poor coverageRvS and DT · D4RL pen-humannormalized scoreImproved
Use TD3+BC on pen-human with only 25 trajectoriesDP methods · D4RL pen-humannormalized scoreHurt
Use DT instead of IQL on pen-humanRCSL vs IQL · D4RL pen-humannormalized scoreImproved
Increase dataset size from pen-human (25 trajectories) to pen-expert (5000 trajectories)DP methods vs RCSL · D4RL pennormalized score: 25 -> 5000 trajectoriesImproved
Use RvS instead of %BC on stitch-easy datasetRvS · Point-mass stitch-easyreturnImproved
Use DT (history-conditioned) instead of RvS on stitch-easyDT · Point-mass stitch-easyreturnHurt

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