Research map / Architectures and efficiency / RNNs and normalization
PDSketch: Integrated Domain Programming, Learning, and Planning
Jiayuan Mao, Tomás Lozano-Pérez, Joshua B. Tenenbaum, Leslie Pack Kaelbling
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 the facing predicate (PDS-Base to PDS-Abs) (ablation) | PDSketch · BabyAI basic | success rate: 0.82 -> 0.99 | Improved |
| add the facing predicate (PDS-Base to PDS-Abs) (ablation) | PDSketch · BabyAI generalization | success rate: 0.62 -> 0.98 | Improved |
| add predefined robot movement rules (PDS-Abs to PDS-Rob) (ablation) | PDSketch · BabyAI generalization | success rate: 0.98 -> 1.00 | Improved |
| add failed demonstrations to Decision Transformer (ablation) | DT · BabyAI basic | success rate: 0.91 -> 0.32 | Hurt |
| add failed demonstrations to Decision Transformer (ablation) | DT · BabyAI generalization | success rate: 0.82 -> 0.19 | Hurt |
| use PDS-Abs instead of model-based DreamerV2 | PDSketch · BabyAI generalization | success rate: 0.79 -> 0.98 | Improved |
| DreamerV2 versus BC on more objects | DreamerV2 · BabyAI generalization | success rate: 0.79 -> 0.79 | No clear effect |
| PDS versus BC with 100 demonstrations | PDSketch · Painting Factory | success rate: 0.70 -> 0.91 | Improved |
| PDS versus DT with 100 demonstrations | PDSketch · Painting Factory | success rate: 0.56 -> 0.91 | Improved |
| increase demonstrations from 100 to 1000 for PDS | PDSketch · Painting Factory | success rate: 0.91 -> 0.99 | Improved |
| PDS versus DT with 1000 demonstrations | PDSketch · Painting Factory | success rate: 0.97 -> 0.99 | Improved |
| use PDSketch-derived heuristic instead of GNN-learned heuristic | A* with PDS-Abs · BabyAI | expanded nodes at 0.8 success: 5 times more efficient | Improved |
Closest papers by meaning
- UniDomain: Pretraining a Unified PDDL Domain from Real-World Demonstrations for Generalizable Robot Task Planning (NeurIPS 2025 · similarity 0.89)
- Learning Domain-Independent Heuristics for Grounded and Lifted Planning (AAAI 2024 · similarity 0.87)
- PO-PDDL: Learning Symbolic POMDPs from Visual Demonstrations for Robot Planning Under Uncertainty (NeurIPS 2026 · similarity 0.86)
- Efficient Robotic Policy Learning via Latent Space Backward Planning (ICML 2025 · similarity 0.85)
- Learning Neuro-Symbolic Abstractions for Robot Planning and Learning (AAAI 2024 · similarity 0.85)
- Leveraging Environment Interaction for Automated PDDL Translation and Planning with Large Language Models (NeurIPS 2024 · similarity 0.85)
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.
