Arm-wise Compositional Generalization in Dual-Arm Vision-Language-Action Models

2026-10-05

AI Summary

This study addresses the limited generalization of skill recomposition in bimanual collaboration by proposing the ACG-Bench benchmark and the AE-VLA model. ACG-Bench establishes the first evaluation standard for bimanual atomic skill recomposition, revealing a complementary mechanism between skill-conditional parameters and attention structures. The proposed AE-VLA integrates vision-language-action models with SkillLoRA adapters, inter-arm attention, and data augmentation strategies, leveraging architectural innovations to enhance generalization on unseen compositional tasks. Experimental results demonstrate that this approach achieves generalization success rates of 21.53% in simulation and 39% on real-world robots, significantly outperforming existing baselines.

Abstract

Generalization in multi-arm collaboration can be studied as composing familiar atomic skills in new ways across arms. However, existing evaluations offer limited insight into which training and architectural choices support this ability under different coordination requirements. We introduce \textbf{ACG-Bench}, a benchmark for \emph{Arm-wise Compositional Generalization} that provides a common testbed for studying skill recomposition in dual-arm policies. It contains 23 task--condition pairs across 8 task families, with 6 in-domain conditions and 17 unseen compositions covering reordering, synchronization, their combination, and cross-task composition. All methods receive the same per-arm atomic prompts, and success requires achieving the task goal while satisfying physical milestones and specified order or timing constraints. Using $\pi_{0.5}$ as a common vision-language-action backbone, we compare representative data-augmentation and architectural strategies with shared source data and a common evaluation protocol. Our architectural study examines arm-token grouping, skill-specific LoRA adapters (SkillLoRA), and arm-wise attention (AWA), highlighting the complementarity of skill-conditioned parameters and attention structure. Combining these choices yields \textbf{AE-VLA}, which achieves 21.53\% generalization success in simulation, compared with 2.94\% for Single $\pi_{0.5}$, 3.06\% for MA-VLA, and 5.53\% for two independently controlled $\pi_{0.5}$ policies. On physical SO101 robots, AE-VLA reaches 39.00\% mean success across five unseen conditions, compared with 10.00\% for the strongest baseline. These findings provide empirical guidance for designing dual-arm policies that generalize beyond fixed training routines.

Problem

• compositional generalization
• dual-arm manipulation
• vision-language-action models
• skill recomposition
• benchmark

Innovation

• Arm-wise Compositional Generalization
• Vision-Language-Action Models
• SkillLoRA
• Arm-wise Attention
• AE-VLA