Research map / Reinforcement learning / Visual and representation RL
Active Hierarchical Exploration with Stable Subgoal Representation 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 |
|---|---|---|---|
| add the potential measure to novelty-based subgoal selection (ablation) | HESS · sparse-reward control tasks | success rate | Helped |
| use cumulative counts instead of immediate counts for novelty (ablation) | HESS · sparse-reward control tasks | sample efficiency | Helped |
| actively select promising subgoals instead of using high-level intrinsic rewards (ablation) | HESS · sparse-reward control tasks | sample efficiency | Helped |
| add stability regularization to the subgoal representation (ablation) | HESS · image-input tasks | success rate | Helped |
| add prioritized sampling to representation learning (ablation) | HESS · image-input tasks | success rate | Helped |
| add stability regularization and prioritized sampling (ablation) | HESS · vector-input tasks | success rate | Mixed |
| extend sampled subgoals to imagined subgoals with positive extended distance (ablation) | HESS · Ant Push and Ant Push (Images) | sample efficiency | Helped |
| increase the subgoal-selection radius instead of perturbing sampled subgoals (ablation) | HESS · Ant Push and Ant Push (Images) | sample efficiency | Hurt |
| use binary rather than continuous stability weights (ablation) | HESS · reported stability-weighting comparison | performance | Helped |
| set representation update interval to 50 episodes rather than a suitable longer interval (ablation) | HESS · Ant Maze and Ant Maze (Images) | learning performance | Hurt |
| set representation update interval to 500 episodes (ablation) | HESS · Ant Maze (Images) | policy learning success | Hurt |
| use a small active-exploration initialization probability of 0.3 (ablation) | HESS · reported initialization comparison | sample efficiency | Hurt |
Closest papers by meaning
- Probabilistic Subgoal Representations for Hierarchical Reinforcement Learning (ICML 2024 · similarity 0.91)
- Diffusion Subgoal Planning for Long-Horizon Offline Goal-Conditioned Reinforcement Learning (NeurIPS 2026 · similarity 0.90)
- Flattening Hierarchies with Policy Bootstrapping (NeurIPS 2025 · similarity 0.89)
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- Hierarchical Reinforcement Learning with Targeted Causal Interventions (ICML 2025 · similarity 0.88)
- Hierarchical Reinforcement Learning with Uncertainty-Guided Diffusional Subgoals (ICML 2025 · 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.