Research map / Reinforcement learning / Visual and representation RL
Hindsight Task Relabelling: Experience Replay for Sparse Reward Meta-RL
Charles Packer, Pieter Abbeel, Joseph E. Gonzalez
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Add hindsight task relabeling during sparse-reward meta-training | PEARL · Sparse-reward goal-reaching environments | average sparse return | Improved |
| Use sparse-reward HTR meta-training instead of dense-reward training | PEARL · Sparse-reward goal-reaching environments | average sparse return | No clear effect |
| Add HTR to PEARL rather than use unmodified sparse-reward PEARL | PEARL · Held-out sparse-reward tasks | average sparse return | Improved |
| Use Episode Clustering instead of Single Episode Relabeling (ablation) | HTR · Point Robot | average return | No clear effect |
| Set relabeling probability K to 0.1 or 0.3 rather than higher tested probabilities (ablation) | HTR · Point Robot | average return | Improved |
| Increase relabeling probability K to 0.5 or 0.8 (ablation) | HTR · Point Robot | mean relabelled task distance: 2.0 -> 0.5 | Hurt |
| Sample context and RL batches from the same relabelled episode (ablation) | HTR with SER · Sparse-reward meta-RL | reward in the RL batch | Improved |
Closest papers by meaning
- Hindsight Foresight Relabeling for Meta-Reinforcement Learning (ICLR 2022 · similarity 0.94)
- Learning Action Translator for Meta Reinforcement Learning on Sparse-Reward Tasks (AAAI 2022 · similarity 0.91)
- MetaCARD: Meta-Reinforcement Learning with Task Uncertainty Feedback via Decoupled Context-Aware Reward and Dynamics Components (AAAI 2024 · similarity 0.90)
- PIPER: Primitive-Informed Preference-based Hierarchical Reinforcement Learning via Hindsight Relabeling (ICML 2024 · similarity 0.88)
- First-Explore, then Exploit: Meta-Learning to Solve Hard Exploration-Exploitation Trade-Offs (NeurIPS 2024 · similarity 0.88)
- Learning Process Rewards via Visitation Matching for Efficient RL (NeurIPS 2026 · similarity 0.88)
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.
