AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

Search on the Replay Buffer: Bridging Planning and Reinforcement Learning

Ben Eysenbach, Ruslan Salakhutdinov, Sergey Levine

NeurIPS 2019 · Reinforcement learning › Visual and representation RL · approach: Goal-conditioned policy learning · 39 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
Add graph search over replay-buffer observations as intermediate waypointsGoal-conditioned RL · 2D navigation with goals over 100 steps awaysuccess rateImproved
Add graph search over replay-buffer observations as intermediate waypointsGoal-conditioned C51 · Visual Navigationsuccess rate: around 90%Improved
Add graph search over replay-buffer observations as intermediate waypointsGoal-conditioned RL · SUNCG held-out houses, goals 10 steps awaysuccess rate: almost 80%; less than 20%Improved
Use SoRB's learned distance predictions instead of SPTM'sVisual navigation graph search · Visual NavigationAUC: 22% higherImproved
Add distributional RL (ablation)SoRB · Visual Navigationsuccess rateImproved
Add ensembles of value functions (ablation)SoRB · Visual Navigation, goals at least 10 steps awaysuccess rate: 10 - 20%Improved
Reduce the replay buffer size by 10x (ablation)SoRB · Visual Navigationsuccess rate: 10xNo clear effect
Use a MaxDist that is too small or too large (ablation)SoRB graph construction · Visual Navigationsuccess rateHurt
Add off-policy goal relabelling (ablation)SoRB non-search policy · Goal-reachingperformanceHurt
Add off-policy goal relabelling (ablation)SoRB search policy · Goal-reachingperformanceNo clear effect
Use search paths to impose lower bounds on Q-value targets (ablation)SoRB training · Goal-reachinglearning speed and convergenceHurt
Use TD3-style aggregation of distributional critics (ablation)SoRB · Goal-reachingperformanceHurt

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