Research map / Reinforcement learning / Visual and representation RL
Towards Improvisational TAMP: Learning Low-Level Shortcuts in Abstract Planning Graphs
Y. Isabel Liu, Bowen Li, Benjamin Eysenbach, Tom Silver
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 learned shortcuts | Pure Planning · Obstacle 2D | plan length: 25.9 -> 17.6 | Improved |
| add learned shortcuts | Pure Planning · Obstacle Tower | plan length: 245.8 -> 79.2 | Improved |
| add learned shortcuts | Pure Planning · Cluttered Drawer | plan length: 352.1 -> 165.8 | Improved |
| add learned shortcuts | Pure Planning · Cleanup Table | plan length: 431.8 -> 115.2 | Improved |
| add abstract planning with shortcuts to pure RL | PPO · Obstacle Tower | success rate: 0% -> 100% | Improved |
| add hierarchical RL access to predefined skills | Pure Planning · Obstacle 2D | plan length: 25.9 -> 25.3 | No clear effect |
| add SOL hierarchical option learning | Pure Planning · Obstacle 2D | plan length: 25.9 -> 24.9 | No clear effect |
| use independent shortcut policies instead of abstract subgoal or abstract HER shared policies (ablation) | SLAP · Four environments | plan length | Improved |
| raise pruning ratio Krollout/Nrollout from 5% to 30% (ablation) | SLAP · Cluttered Drawer | plan length: 178.2 -> 333.0 | Hurt |
| raise pruning ratio from 5% to 10% (ablation) | SLAP · Cluttered Drawer | plan length: 178.2 -> 167.0 | No clear effect |
| use SAC instead of PPO as shortcut RL backbone (ablation) | SLAP · Cleanup Table | plan length: 113.7 -> 160.2 | Hurt |
| use SAC+HER instead of PPO as shortcut RL backbone (ablation) | SLAP · Cleanup Table | plan length: 113.7 -> 131.0 | Mixed |
Closest papers by meaning
- Efficient Robotic Policy Learning via Latent Space Backward Planning (ICML 2025 · similarity 0.88)
- Autonomous Option Invention for Continual Hierarchical Reinforcement Learning and Planning (AAAI 2025 · similarity 0.87)
- Learning Planning Abstractions from Language (ICLR 2024 · similarity 0.87)
- Overcoming Slow Decision Frequencies in Continuous Control: Model-Based Sequence Reinforcement Learning for Model-Free Control (ICLR 2025 · similarity 0.87)
- Consciousness-Inspired Spatio-Temporal Abstractions for Better Generalization in Reinforcement Learning (ICLR 2024 · similarity 0.87)
- Scalable Decision-Making in Stochastic Environments through Learned Temporal Abstraction (ICLR 2025 · 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.
