Research map / Reinforcement learning / Visual and representation RL
Learning Space Partitions for Path Planning
Kevin Yang, Tianjun Zhang, Chris Cummins, Brandon Cui, Benoit Steiner, Linnan Wang, Joseph E. Gonzalez, Dan Klein, Yuandong Tian
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 reward-informed learned partitions and latent-space planning | LaP^3 · MiniWorld MazeS3 | success percentage: LaP^3: 57.0 $\pm$ 3.1; VOOT: 26.2 $\pm$ 2.7 | Improved |
| Use reward-informed learned partitions and latent-space planning | LaP^3 · MiniWorld FourRooms | success percentage: LaP^3: 89.1 $\pm$ 2.0; CMA-ES: 79.7 $\pm$ 2.5 | Improved |
| Use reward-informed learned partitions and latent-space planning | LaP^3 · MiniWorld SelectObj | success percentage: LaP^3: 45.3 $\pm$ 3.1; PPO: 31.3 $\pm$ 8.2 | Improved |
| Replace the PETS CEM planner with LaP^3 | PETS · MiniWorld FourRooms | fraction of environment seeds solved: 66.9 $\pm$ 6.3 -> 83.1 $\pm$ 2.3 | Improved |
| Replace the PETS CEM planner with LaP^3 | PETS · MiniWorld SelectObj | fraction of environment seeds solved: 7.2 $\pm$ 1.9 -> 19.4 $\pm$ 1.8 | Improved |
| Use reward-informed learned partitions and latent-space planning | LaP^3 · MiniGrid MR-N4S5 | reward: -14.31$\pm$0.78 -> -11.68$\pm$0.75 | Improved |
| Use reward-informed learned partitions and latent-space planning | LaP^3 · Molecular design DRD2 | max property value: 0.323 $\pm$ 0.016 -> 0.648 $\pm$ 0.026 | Improved |
| Use mean rather than maximum node value (ablation) | LaP^3 · MiniWorld MazeS3 | success percentage: 57.0 $\pm$ 3.1 -> 43.4 $\pm$ 3.1 | Hurt |
| Remove latent partition space (ablation) | LaP^3 · MiniWorld MazeS3 | success percentage: 57.0 $\pm$ 3.1 -> 26.6 $\pm$ 2.8 | Hurt |
| Remove tree traversal and directly select the highest-UCB leaf (ablation) | LaP^3 · MiniWorld MazeS3 | success percentage: 57.0 $\pm$ 3.1 -> 24.6 $\pm$ 2.7 | Hurt |
| Remove the UCB exploration bonus (ablation) | LaP^3 · MiniWorld MazeS3 | success percentage: 57.0 $\pm$ 3.1 -> 28.1 $\pm$ 2.8 | Hurt |
| Use data-driven reward-sensitive partitioning instead of random partitioning | LaP^3 · MiniWorld | estimated average Lipschitz constant and c_k | Improved |
Closest papers by meaning
- CompassNav: Steering From Path Imitation to Decision Understanding In Navigation (ICLR 2026 · similarity 0.86)
- Accelerating Motion Planning via Optimal Transport (NeurIPS 2023 · similarity 0.85)
- Neural Neighborhood Search for Multi-agent Path Finding (ICLR 2024 · similarity 0.85)
- C-Planning: An Automatic Curriculum for Learning Goal-Reaching Tasks (ICLR 2022 · similarity 0.85)
- Learn to Follow: Decentralized Lifelong Multi-Agent Pathfinding via Planning and Learning (AAAI 2024 · similarity 0.84)
- STRIDER: Navigation via Instruction-Aligned Structural Decision Space Optimization (NeurIPS 2025 · similarity 0.84)
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.
