Research map / Reinforcement learning / Visual and representation RL
Successor Feature Landmarks for Long-Horizon Goal-Conditioned Reinforcement Learning
Christopher Hoang, Sungryull Sohn, Jongwook Choi, Wilka Torrico Carvalho, Honglak Lee
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 |
|---|---|---|---|
| Use SFS instead of SPTM's reachability network as the distance metric (ablation) | SFL · ViZDoom SGM-Map, random spawn | success rate: 83 ± 1.2% -> 92 ± 1.7%; 57 ± 2.6% -> 82 ± 0.6%; 24 ± 1.9% -> 67 ± 2.6% | Improved |
| Sample frontier landmarks by inverse visitation count instead of uniformly at random (ablation) | SFL · ViZDoom Test-1 | state coverage: 72.3 ± 1.90% -> 79.4 ± 0.65% | Improved |
| Build landmarks progressively and retain landmark metadata instead of using clustering (ablation) | SFL · Three-room MultiRoom | success rate: 35.6 ± 18.8% -> 75.0 ± 14.6% | Improved |
| Build landmarks progressively rather than with clustering (ablation) | SFL · FourRoom | average pairwise geodesic distance between landmarks: 5.72 ± 0.40 -> 6.72 ± 0.43 | Improved |
| Use SFL rather than the strongest reported baseline | Long-horizon goal-reaching · ViZDoom random spawn, hard tasks | success rate: SGM-Map: 26% -> 67%; Test-2: 32% -> 48%; Test-6: 27% -> 60% | Improved |
| Use SFL rather than the strongest reported baseline | Exploration and goal-reaching · ViZDoom fixed spawn | success rate: Test-1 Hard: 1% -> 62%; Test-1 Hardest: 1% -> 50%; Test-4 Hard: 7% -> 27%; Test-4 Hardest: 7% -> 23% | Improved |
Closest papers by meaning
- Learning Temporal Distances: Contrastive Successor Features Can Provide a Metric Structure for Decision-Making (ICML 2024 · similarity 0.88)
- Flattening Hierarchies with Policy Bootstrapping (NeurIPS 2025 · similarity 0.87)
- Exploring the Edges of Latent State Clusters for Goal-Conditioned Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Offline Goal-conditioned Reinforcement Learning with Quasimetric Representations (NeurIPS 2025 · similarity 0.87)
- Reward-Aware Proto-Representations in Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Diffusion Subgoal Planning for Long-Horizon Offline Goal-Conditioned Reinforcement Learning (NeurIPS 2026 · 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.
