CoRMA: Contrastive RMA for Contact-Rich Meta-Adaptation

2026-05-21

AI Summary

This work addresses the performance degradation of simulation-to-reality transfer in force-dominated assembly tasks, particularly when target poses are corrupted by noise, leading to a sharp drop in success rates. To tackle this challenge, the authors propose a semantic contact context–based meta-adaptation framework that leverages a causal Transformer to online fuse force, proprioceptive, and action history signals. This enables unsupervised inference of compact 6D contact semantics, which serve as a reusable adaptation interface. The approach facilitates real-time policy adaptation without requiring demonstrations, privileged information, or gradient updates. Integrated with contrastive learning and semantic regression, the method significantly outperforms baselines such as FORGE on PegInsert, GearMesh, and NutThread tasks, demonstrating substantially improved real-world success rates, robustness, and transferability.

Abstract

We present CoRMA(Contrastive Robotic Motor Adaptation), a context-based meta-adaptation framework that modifies RMA for force-dominant assembly. CoRMA replaces raw simulator-parameter adaptation with a compact 6D simulator-only semantic contact context describing contact onset, lateral engagement, guided transition, contact direction, and jamming. A deployable causal Transformer adapter infers this context online from force, proprioceptive, and action histories using semantic regression and a force-regime contrastive objective. At deployment, oracle context is removed and replaced by the inferred context, enabling within-episode adaptation without demonstrations, privileged inputs, or gradient updates. We evaluate CoRMA on PegInsert, GearMesh, and NutThread in Isaac Lab / Isaac Sim~5.0 and on a real Marvin arm. Compared with FORGE baselines that achieve high simulation success but degrade substantially on hardware, CoRMA retains higher verified real success under controlled target-pose noise. These results support semantic contact inference as a reusable adaptation interface within a related assembly task family, while broader unseen-task generalization and Real2Sim calibration remain future work.

Problem

• contact-rich manipulation
• meta-adaptation
• sim-to-real transfer
• robotic assembly
• force-based adaptation

Innovation

• Contrastive Learning
• Semantic Contact Context
• Meta-Adaptation
• Force-Dominant Assembly
• Causal Transformer