Learning the Contact Manifold for Accurate Pose Estimation During Peg-in-Hole Insertion of Complex Geometries

2025-05-25

AI Summary

Accurate estimation of contact states and SE(3) poses remains challenging for complex non-convex parts during tight-tolerance (0.1–1.0 mm) shaft-hole assembly. Method: This paper introduces the first lightweight hybrid pose estimation algorithm based on contact manifold alignment. Relying solely on sparse tactile contact observations, it constructs a local contact submanifold within 10 seconds via probe motion and aligns it with a precomputed offline contact manifold—eliminating conventional k-NN search in favor of a dedicated neural projection network for manifold alignment. Contribution/Results: The proposed method achieves a 95× speedup and an 18% improvement in pose accuracy over k-NN baselines. Evaluated on three industrial part classes, it attains a 93.3% insertion success rate—4.1× higher than a state-unaware baseline—and delivers sub-millimeter positional and sub-degree rotational accuracy.

Technology Category

• Intelligent Robots: State Estimation
• Machine Learning: Learning with Manifolds
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Contact-rich assembly of complex, non-convex parts with tight tolerances remains a formidable challenge. Purely model-based methods struggle with discontinuous contact dynamics, while model-free methods require vast data and often lack precision. In this work, we introduce a hybrid framework that uses only contact-state information between a complex peg and its mating hole to recover the full SE(3) pose during assembly. In under 10 seconds of online execution, a sequence of primitive probing motions constructs a local contact submanifold, which is then aligned to a precomputed offline contact manifold to yield sub-mm and sub-degree pose estimates. To eliminate costly k-NN searches, we train a lightweight network that projects sparse contact observations onto the contact manifold and is 95x faster and 18% more accurate. Our method, evaluated on three industrially relevant geometries with clearances of 0.1-1.0 mm, achieves a success rate of 93.3%, a 4.1x improvement compared to primitive-only strategies without state estimation.

Problem

• Accurate pose estimation for peg-in-hole insertion of complex geometries
• Overcoming discontinuous contact dynamics in model-based methods
• Reducing data needs and improving precision in model-free methods

Innovation

• Hybrid framework using contact-state information
• Local contact submanifold alignment for pose estimation
• Lightweight network for fast contact manifold projection