AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

Generalized Hindsight for Reinforcement Learning

Alexander Li, Lerrel Pinto, Pieter Abbeel

NeurIPS 2020 · Reinforcement learning › Visual and representation RL · approach: Probabilistic and autonomous RL · 20 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
Use AIR task relabeling instead of no relabelingSAC multi-task RL · PointTrajectory, PointReacher, Fetch, HalfCheetahMultiObjective, AntDirectionlearning performanceImproved
Use Advantage relabeling instead of no relabelingSAC multi-task RL · PointTrajectory, PointReacher, Fetch, HalfCheetahMultiObjective, AntDirectionlearning performanceImproved
Select a few tasks by AIR rather than relabeling with random tasksSAC multi-task RL · PointTrajectory, PointReacher, Fetch, HalfCheetahMultiObjective, AntDirectionlearning performanceImproved
Use AIR instead of HERSAC multi-task RL · PointReacher and Fetchasymptotic performanceImproved
Replace AIR percentile-based task selection with Advantage relabeling (ablation)Generalized Hindsight · PointTrajectory, PointReacher, Fetch, HalfCheetahMultiObjective, AntDirectionlearning performanceNo clear effect
Choose relabels by maximum trajectory reward instead of AIR percentile (ablation)Generalized Hindsight · PointReacher relabeling fidelityrelabeling fidelityHurt

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