Manifold-Guided Motion Planning for Tight Assemblies

· Editorial Team estimated
motion planning tight assembly RRT manifold learning contact constraints

CMG-RRT proposes critical-manifold guided sampling-based planning, achieving 100% success on all tight assembly benchmarks including the first automatic solution of the Elk disentanglement puzzle, with a probabilistic completeness proof.

Paper · arXiv:2607.17898

Background

Tight assembly poses a fundamental challenge in robotics: parts must mate with near-zero clearance, making it extremely difficult to find feasible paths from initial poses to assembled configurations. In such scenarios, valid solutions lie on or near the “critical manifold” — the subset of configuration space where parts are in contact. Existing planners, however, lack explicit exploitation of this geometric structure.

Core Innovation

This paper introduces Critical-Manifold Guided RRT (CMG-RRT), a sampling-based planner designed specifically for tight assembly problems. The key insight is that valid solution paths in tight assemblies must lie on or near the critical manifold. CMG-RRT guides exploration by adaptively biasing sampling toward neighborhoods of this manifold using a hierarchical subdivision of configuration space.

The paper also provides a probabilistic completeness proof for CMG-RRT under standard clearance assumptions, establishing a theoretical guarantee of reliability.

Results

CMG-RRT achieves 100% success rate across all tested rotational assembly benchmarks, including — to the best of our knowledge — the first fully automatic solution of the Elk disentanglement puzzle. This previously unsolved puzzle validates the algorithm’s ability to handle extreme tight-clearance scenarios. The software is open-source.

Limitations

The method is validated primarily on desk-level rigid-body assemblies. Compliant assembly, part deformation, and more complex multi-part sequences remain to be explored. The probabilistic completeness proof relies on standard clearance assumptions; behavior under true zero-clearance conditions requires further analysis.

Industry Impact

Tight assembly is a critical bottleneck in electronics manufacturing, automotive, and aerospace assembly lines. CMG-RRT’s 100% success on extreme-tolerance scenarios with theoretical guarantees has immediate application potential for industrial robot programming, automated fixture design, and high-precision low-volume assembly.