Research map / Reinforcement learning / Visual and representation RL
Maximum Entropy Gain Exploration for Long Horizon Multi-goal Reinforcement Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| select goals from low-density regions of the historical achieved-goal distribution | MEGA and OMEGA · PointMaze | sample efficiency: almost 100 times faster | Helped |
| select goals from low-density regions of the historical achieved-goal distribution | MEGA and OMEGA · AntMaze | sample efficiency: roughly 10 times faster | Helped |
| select goals from low-density regions of the historical achieved-goal distribution | MEGA and OMEGA · long-horizon multi-goal tasks | sample efficiency: more than an order of magnitude better sample efficiency | Helped |
| use desired-goal sampling rather than frontier-focused intrinsic goals | HER · PointMaze | success | Hurt |
| sample goals from the achieved-goal distribution rather than its low-density frontier | Achieved · PointMaze | success | Hurt |
| sample diverse achieved goals rather than low-density frontier goals | Diverse · PointMaze | exploration progress | Hurt |
| use intermediate-difficulty goal selection | GoalDisc · AntMaze | success | Helped |
| add MEP experience reweighting to HER (ablation) | HER · FetchPickAndPlace | success | Helped |
| add MEP experience reweighting to HER (ablation) | HER · PointMaze | success | No clear effect |
| add MEP experience reweighting (ablation) | MEGA and OMEGA · PointMaze and FetchPickAndPlace | performance | No clear effect |
| anneal intrinsic goal selection toward the desired-goal distribution (ablation) | OMEGA · FetchStack2 | long-run stability | Helped |
| use minimum-density goal selection instead of intermediate-difficulty selection | MEGA · AntMaze | performance | No clear effect |
Closest papers by meaning
- Maximum Entropy-Regularized Multi-Goal Reinforcement Learning (ICML 2019 · similarity 0.91)
- Task-Agnostic Exploration via Policy Gradient of a Non-Parametric State Entropy Estimate (AAAI 2021 · similarity 0.90)
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- f-Policy Gradients: A General Framework for Goal-Conditioned RL using f-Divergences (NeurIPS 2023 · similarity 0.88)
- Towards Principled Unsupervised Multi-Agent Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Exploring the Edges of Latent State Clusters for Goal-Conditioned Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
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.