Research map / Reinforcement learning / Visual and representation RL
Landmark-Guided Subgoal Generation in Hierarchical Reinforcement Learning
Junsu Kim, Younggyo Seo, Jinwoo Shin
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 |
|---|---|---|---|
| Guide high-level subgoals toward landmarks sampled for coverage and novelty | HIGL versus HRAC · Ant Maze (U-shape, sparse), 10 × 10^5 timesteps | success rate: 17.6% -> 65.1% | Improved |
| Add coverage-based landmarks (ablation) | HIGL · Ant Maze (U-shape, dense) | performance | Improved |
| Add novelty-based landmarks (ablation) | HIGL with coverage-based sampling · Ant Maze (U-shape, dense) | performance | Improved |
| Use pseudo-landmarks instead of raw selected landmarks (ablation) | HIGL · Ant Maze (W-shape, dense) | performance | Improved |
| Increase the number of landmarks (ablation) | HIGL · Ant Maze (U-shape, dense) | performance | Improved |
| Set pseudo-landmark shift magnitude too small (ablation) | HIGL · Ant Maze (U-shape, dense) | performance | Hurt |
| Set pseudo-landmark shift magnitude to 4 (ablation) | HIGL · Ant Maze (U-shape, dense) | performance | Hurt |
| Set adjacency degree too large (k=20) (ablation) | HIGL · Ant Maze (U-shape, dense) | performance | Hurt |
| Use shortest-transition-distance-based queue discarding instead of L2-distance-based discarding (ablation) | HIGL novelty priority queue · Additional experiments | performance | Mixed |
| Set shift magnitude automatically from average selected-landmark distance (ablation) | HIGL · Additional experiments | performance | Improved |
| Use HIGL rather than HRAC | Ant Maze (U-shape), 24 × 24, extended timestep · Ant Maze (U-shape), 24 × 24 | training time: 6 hours -> 13 hours | Hurt |
Closest papers by meaning
- Successor Feature Landmarks for Long-Horizon Goal-Conditioned Reinforcement Learning (NeurIPS 2021 · similarity 0.91)
- Flattening Hierarchies with Policy Bootstrapping (NeurIPS 2025 · similarity 0.89)
- Hierarchical Reinforcement Learning with Targeted Causal Interventions (ICML 2025 · similarity 0.89)
- Diffusion Subgoal Planning for Long-Horizon Offline Goal-Conditioned Reinforcement Learning (NeurIPS 2026 · similarity 0.88)
- Probabilistic Subgoal Representations for Hierarchical Reinforcement Learning (ICML 2024 · similarity 0.88)
- Reconciling Spatial and Temporal Abstractions for Goal Representation (ICLR 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.
