Title: TA-VLA: Elucidating the Design Space of Torque-aware Vision-Language-Action Models Show Chinese title

Title: TA-VLA：阐明扭矩感知视觉语言动作模型的设计空间

Abstract: Many robotic manipulation tasks require sensing and responding to force signals such as torque to assess whether the task has been successfully completed and to enable closed-loop control. However, current Vision-Language-Action (VLA) models lack the ability to integrate such subtle physical feedback. In this work, we explore Torque-aware VLA models, aiming to bridge this gap by systematically studying the design space for incorporating torque signals into existing VLA architectures. We identify and evaluate several strategies, leading to three key findings. First, introducing torque adapters into the decoder consistently outperforms inserting them into the encoder.Third, inspired by joint prediction and planning paradigms in autonomous driving, we propose predicting torque as an auxiliary output, which further improves performance. This strategy encourages the model to build a physically grounded internal representation of interaction dynamics. Extensive quantitative and qualitative experiments across contact-rich manipulation benchmarks validate our findings. Show Chinese abstract

Abstract: 许多机器人操作任务需要感知和响应力信号，如扭矩，以评估任务是否成功完成并实现闭环控制。 然而，当前的视觉-语言-动作（VLA）模型缺乏整合此类细微物理反馈的能力。 在本工作中，我们探索了扭矩感知的VLA模型，旨在通过系统研究将扭矩信号融入现有VLA架构的设计空间来弥合这一差距。 我们识别并评估了几种策略，得出三个关键发现。 首先，在解码器中引入扭矩适配器始终优于在编码器中插入它们。第三，受自动驾驶中联合预测和规划范式的启发，我们提出将扭矩作为辅助输出进行预测，这进一步提高了性能。 该策略鼓励模型构建一个物理基础的交互动力学内部表示。 在接触丰富的操作基准上的大量定量和定性实验验证了我们的发现。