Research map / Vision-language and multimodal / Robot action and navigation
Decompose and Recompose: Reasoning New Skills from Existing Abilities for Cross-Task Robotic Manipulation
Xitie Zhang, Aming WU, Yahong Han
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 dynamic demonstrations library (visual-semantic and skill-sequence retrieval) (ablation) | random-demo baseline · AGNOSTOS (ablation) | overall success rate (%): 21.6 -> 23.3 | Improved |
| add coverage-aware demonstrations library (ablation) | BS + DDL · AGNOSTOS (ablation) | overall success rate (%): 23.3 -> 24.9 | Improved |
| add skill-augmented in-context learning (ablation) | BS + DDL + CDL · AGNOSTOS (ablation) | overall success rate (%): 24.9 -> 26.4 | Improved |
| use DINOv3 instead of CLIP visual encoder (ablation) | Decompose and Recompose · AGNOSTOS (ablation) | overall success rate (%): 24.8 -> 26.4 | Improved |
| use DINOv3 instead of DINOv2 visual encoder (ablation) | Decompose and Recompose · AGNOSTOS (ablation) | overall success rate (%): 25.7 -> 26.4 | Improved |
| use Qwen2.5-7B instead of Llama3.0-8B as LLM (ablation) | Decompose and Recompose · AGNOSTOS (ablation) | overall success rate (%): 22.4 -> 26.4 | Improved |
| use Qwen2.5-7B instead of Ministral-8B as LLM (ablation) | Decompose and Recompose · AGNOSTOS (ablation) | overall success rate (%): 20.2 -> 26.4 | Improved |
| use Qwen2.5-7B instead of InternLM3-8B as LLM (ablation) | Decompose and Recompose · AGNOSTOS (ablation) | overall success rate (%): 24.1 -> 26.4 | Improved |
| increase in-context demonstrations from 5 to 20 (ablation) | Decompose and Recompose · AGNOSTOS (ablation) | overall success rate (%): 19.8 -> 26.4 | Improved |
| provide in-context demonstrations (0 vs 5) (ablation) | Decompose and Recompose · AGNOSTOS (ablation) | overall success rate (%): 0.0 -> 19.8 | Improved |
| increase coverage-aware demonstrations from 0 to 3 (ablation) | Decompose and Recompose · AGNOSTOS (ablation) | overall success rate (%): 24.9 -> 26.4 | Improved |
| increase coverage-aware demonstrations from 3 to 4 (ablation) | Decompose and Recompose · AGNOSTOS (ablation) | overall success rate (%): 26.4 -> 26.1 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SAM2Act — SAM2Act: Integrating Visual Foundation Model with A Memory Architecture for Robotic Manipulation (ICML 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SkillDiffuser: Interpretable Hierarchical Planning via Skill Abstractions in Diffusion-Based Task Execution (CVPR 2024 · similarity 0.90)
- UniDomain: Pretraining a Unified PDDL Domain from Real-World Demonstrations for Generalizable Robot Task Planning (NeurIPS 2025 · similarity 0.89)
- Exploring the Limits of Vision-Language-Action Manipulation in Cross-task Generalization (NeurIPS 2025 · similarity 0.89)
- RDD: Retrieval-Based Demonstration Decomposer for Planner Alignment in Long-Horizon Tasks (NeurIPS 2025 · similarity 0.88)
- GravMAD: Grounded Spatial Value Maps Guided Action Diffusion for Generalized 3D Manipulation (ICLR 2025 · similarity 0.88)
- InfoCon: Concept Discovery with Generative and Discriminative Informativeness (ICLR 2024 · similarity 0.88)
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.
