Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces ACG-Bench, a new benchmark for evaluating arm-wise compositional generalization in dual-arm vision-language-action (VLA) models, comprising 23 task-condition pairs across 8 task families with novel compositions. Through empirical studies on a π_{0.5} backbone, the research compares data augmentation and architectural strategies, finding that skill-conditioned parameters (SkillLoRA) and arm-wise attention (AWA) are complementary. The proposed AE-VLA architecture achieves significant generalization improvements, reaching 21.53% in simulation and 39.00% on physical robots, outperforming baselines by a substantial margin.
Dual-arm robots can generalize to unseen task compositions by combining skill-conditioned parameters and arm-wise attention, achieving 39% success on physical hardware.
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 ACG-Bench, a benchmark for 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 π_{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 AE-VLA, which achieves 21.53\% generalization success in simulation, compared with 2.94\% for Single π_{0.5}, 3.06\% for MA-VLA, and 5.53\% for two independently controlled π_{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.