AI Paper Atlas

Research map / Reinforcement learning / Offline RL

Improving Zero-Shot Offline RL via Behavioral Task Sampling

Nazim Bendib, Nicolas Perrin-Gilbert, Olivier Sigaud

ICML 2026 · Reinforcement learning › Offline RL · approach: Offline-to-online fine-tuning · 0 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
replace uniform hypersphere task sampling with Behavioral Task Distribution (GMM over dataset-derived task vectors)SF and FB zero-shot RL (AEnc, Trans, LRA-P, LRA-SR, BYOL, BYOL-gamma, FB) · ExORL Cheetah, Walker, Quadrupedzero-shot test performance: average +20%; better in 13 of 15 tests, comparable in 2Improved
use BTD-sampling with FBForward-Backward representation · Cheetah, Walker, Quadrupedzero-shot test performanceImproved
use BTD-sampling instead of uniform samplingall baseline methods · Cheetah, Walker, Quadrupedvariance across representation methodsImproved
increase latent task dimension from 50 to 1000 with uniform sampling (ablation)baseline uniform task sampling · Cheetah, Walker, Quadrupedtest performanceHurt
increase latent task dimension from 50 to 1000 with BTD-sampling (ablation)BTD-sampling · Cheetah, Walker, Quadrupedtest performanceNo clear effect
mix uniform sampled tasks into BTD-sampling (interpolation alpha from 0 to 1) (ablation)BTD-sampling · Cheetah, Walker, Quadrupedtest performanceHurt
mix uniform tasks into BTD-sampling (ablation)BTD-sampling · Quadrupedtest performanceNo clear effect

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