Research Analyzer
← Back ICRA 2026

CG-THWM: Curriculum-Guided Temporal Haptic World Modeling for Peg-In-Hole Tasks

Xinli Zhong, Feng Han, Manya Xu, Mu Li, Daqiang Zhang, Jianwei Niu

PDF

AI summary

Key figure (auto-extracted from paper)
CG-THWM achieves 100% success on standard peg-in-hole baselines and 70% in complex scenarios by fusing temporal haptic signals with a curriculum-guided world model for robust contact-rich manipulation.
peg-in-hole haptic world model curriculum learning contact-rich manipulation model-based RL robotic assembly

Problem

Complex peg-in-hole tasks involve contact-rich, nonsmooth dynamics with irregular geometries and tight clearances, causing traditional model-free RL and heuristic methods to fail due to poor generalization, sample inefficiency, and lack of standardized evaluation datasets.

Approach

The method aligns proprioceptive states with temporal haptic encoders in a latent space, uses haptic-aware attention to capture contact events, and trains via a staged contact-geometry curriculum to stabilize learning and improve long-horizon planning.

Key results

  • 100% success rate on standard simulation baselines
  • 70% mean success rate in complex, failure-prone scenarios
  • Release of the ComplexPeg-Hole dataset with 100,000 diverse configurations
  • Curriculum-guided training enables stable convergence and robust generalization

Why it matters

Provides a robust, data-driven framework for precision assembly that reduces reliance on manual tuning and heuristic search, benefiting industrial automation and service robotics.

Abstract

Fine-tolerance peg-in-hole manipulation demands high precision under contact-rich, nonsmooth dynamics, where irregular geometries, inclinations, and tight-clearance inter- ference often cause model-free reinforcement learning (RL) to fail. We propose the Curriculum-Guided Temporal Haptic World Model (CG-THWM), which couples a world model with temporal haptic information and trains it via a staged curriculum. The world model supports efficient long-horizon planning with value estimation, while temporal haptic signals expose critical contact events; the curriculum stabilizes training and improves generalization. To enable rigorous evaluation, we construct a dataset for complex insertions that covers irregular, inclined, and interference-rich settings. In simulation, CG- THWM attains a 100% success rate on standard baselines and a 70% mean success rate in scenarios where conventional RL fails. These results highlight CG-THWM’s potential for industrial and service applications.

Index terms

Assembly Reinforcement Learning Manipulation Planning

Related papers