Research map / 3D vision / 3D generation
State-Based Disassembly Planning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes State-Based Disassembly Planning (SBDP), a state-storing search that prioritizes translational over rotational physics-based simulation and uses Directional Blocking Graph evaluation functions and constraints to cut simulations.
Problem
Physics-based disassembly planning (PDP) is inefficient because exhaustive sequence and path search causes redundant simulations, regenerates trajectories, and requires users to choose translational or rotational motion.
Approach
Stores visited and collision states in open/collision lists so search resumes from intermediate states, alternates translational and rotational planning automatically, and guides/prunes search with two DBG-derived evaluation functions (f_c, f_a) and DBG action constraints built via SDF-based static analysis and simulation feedback.
Key results
- SBDP$^*$ solves 4135 of 4196 problems (98.55%), more than SBDP (4106), PDP$^*$ (3976), PDP$_r$ (3961) and PDP$_t$ (3772).
- On commonly solved problems, SBDP$^*$ reduces average simulations by 55%/80%/75% versus SBDP, PDP$^*$, and PDP$_r$.
- SBDP$^*$ reduces average path planning time by 40%/50%/70% and disassembly planning time by 20%/35%/50% relative to SBDP, PDP$^*$, and PDP$_r$.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use state-based search instead of PDP search | SBDP vs PDP* · Industrial assemblies, all (4196) | problems solved: 4106 vs 3976 | Helped |
| add DBG-derived evaluation functions and DBG constraints (ablation) | SBDP · Industrial assemblies, all (4196) | problems solved: 4106 -> 4135 | Helped |
| add DBG-derived evaluation functions and DBG constraints (ablation) | SBDP · Industrial assemblies, all commonly solved (3954) | average simulations: 65 -> 29 | Helped |
| add DBG-derived evaluation functions and DBG constraints (ablation) | SBDP · Industrial assemblies, all commonly solved (3954) | average path planning time: 78 -> 47 | Helped |
| add DBG-derived evaluation functions and DBG constraints (ablation) | SBDP · Industrial assemblies, all commonly solved (3954) | average disassembly planning time: 134 -> 111 | Helped |
| use state-based search rather than PDP* search | SBDP* vs PDP* · Industrial assemblies, all commonly solved (3954) | average simulations: 159 -> 29 | Helped |
| use state-based search rather than PDP* search | SBDP* vs PDP* · Industrial assemblies, all commonly solved (3954) | average path planning time: 97 -> 47 | Helped |
| use state-based search rather than PDP* search | SBDP* vs PDP* · Industrial assemblies, all commonly solved (3954) | average disassembly planning time: 166 -> 111 | Helped |
| use DBG-guided SBDP* rather than SBDP (ablation) | SBDP* vs SBDP · Industrial assemblies, all (4196) | problems solved: -36/+7 | Mixed |
Limitations
- Occasional incorrect collision detection yields penetrating disassembly trajectories for some problems.
- Rotational planning is computationally demanding and rotational static analysis is omitted.
- Applying classical planners with heuristics to SBDP is challenging because effects are only known after simulation.
Experimental setup
Datasets: large-scale industrial assembly dataset (thousands of physically valid assemblies)
Benchmarks: industrial assembly disassembly benchmark (4196 problems: small, medium, large)
Metrics: number of problems solved / success rate number of simulations path planning time disassembly planning time
Baselines: PDP$^*$ PDP$_t$ PDP$_r$ SBDP (without DBG heuristics)
Builds on: PDP (physics-based disassembly planning) Directional Blocking Graph
Closest papers by meaning
- SRSA: Skill Retrieval and Adaptation for Robotic Assembly Tasks (ICLR 2025 · similarity 0.81)
- BaB-ND: Long-Horizon Motion Planning with Branch-and-Bound and Neural Dynamics (ICLR 2025 · similarity 0.80)
- RobotSmith: Generative Robotic Tool Design for Acquisition of Complex Manipulation Skills (NeurIPS 2025 · similarity 0.80)
- GENMANIP: LLM-driven Simulation for Generalizable Instruction-Following Manipulation (CVPR 2025 · similarity 0.80)
- Grasp-Then-Plan with Failure Attribution: A Closed Two-Stage Framework for Precise and Generalizable Robotic Manipulation (NeurIPS 2026 · similarity 0.79)
- Position: Scaling Simulation is Neither Necessary Nor Sufficient for In-the-Wild Robot Manipulation (ICML 2024 · similarity 0.79)
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.