Research Analyzer
← Back ICRA 2026

Connectivity-Aware Representations for Constrained Motion Planning Via Multi-Scale Contrastive Learning

Suhyun Jeon, Yumin Lim, Woo-Jeong Baek, Hyeonseo Kim, Suhan Park, Jaeheung Park

PDF

AI summary

Key figure (auto-extracted from paper)
Learning connectivity-aware representations via multi-scale contrastive learning significantly boosts constrained motion planning success rates while cutting planning time.
constrained motion planning connectivity-aware representation multi-scale contrastive learning manifold learning inverse kinematics robotic manipulation

Problem

Constrained motion planning often fails or becomes inefficient when start and goal configurations lie in disconnected regions or narrow passages, especially in redundant manipulators with multiple inverse kinematics solutions. Existing methods lack explicit connectivity awareness to guide configuration selection prior to planning.

Approach

The method generates multi-scale pseudo-labels by applying manifold learning and clustering across different neighborhood scales to capture local and global connectivity. These labels supervise a contrastive learning framework that embeds reachable configurations closer together in a latent space, enabling pre-planning selection of connected start-goal pairs.

Key results

  • Multi-scale pseudo-label generation via UMAP and HDBSCAN clustering
  • Connectivity-aware latent representation via multi-scale contrastive learning
  • 1.9× higher success rate in constrained manipulation tasks
  • 0.43× reduction in planning time compared to baselines

Why it matters

Enables robots to reliably perform complex, constraint-heavy manipulation tasks by intelligently selecting reachable start-goal configurations before planning.

Abstract

The objective of constrained motion planning is to connect start and goal configurations while satisfying task- specific constraints. Motion planning becomes inefficient or infeasible when the configurations lie in disconnected regions, known as essentially mutually disconnected (EMD) compo- nents. Constraints further restrict feasible space to a lower- dimensional submanifold, while redundancy introduces addi- tional complexity because a single end-effector pose admits infinitely many inverse kinematic solutions that may form discrete self-motion manifolds. This paper addresses these challenges by learning a connectivity-aware representation for selecting start and goal configurations prior to planning. Joint configurations are embedded into a latent space through multi- scale manifold learning across neighborhood ranges from local to global, and clustering generates pseudo-labels that supervise a contrastive learning framework. The proposed framework provides a connectivity-aware measure that biases the selection of start and goal configurations in connected regions, avoiding EMDs and yielding higher success rates with reduced planning time. Experiments on various manipulation tasks showed that our method achieves 1.9 times higher success rates and reduces the planning time by a factor of 0.43 compared to baselines.

Index terms

Constrained Motion Planning Bimanual Manipulation Representation Learning

Related papers