发表机构
Tsinghua University; Shenzhen International Graduate School; Nanyang Technological University; X Square Robot; Xspark AI（清华大学; 深圳国际研究生院; 南洋理工大学; X Square机器人公司; Xspark人工智能公司）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
UniReflex是一种即插即用框架，通过快速反射网络和解耦的可变阻抗控制，为冻结的生成策略提升接触稳定性与成功率，同时保留位置精度且延迟大幅降低。
AI 中文摘要
生成式模仿学习策略在轨迹规划方面表现出色，但缺乏闭环力调节能力，而直接融入力模态通常需要重新设计或重新训练网络。我们提出UniReflex，这是一个通用即插即用框架，可在无需进一步慢主干微调的情况下，为冻结的生成策略配备用于接触调节的可变阻抗控制（VIC），该控制由演示过程中收集的力方向意图引导。通过非侵入式地拦截来自动作头的深度潜在表示，UniReflex驱动一个快速反射网络，该网络将主动力施加与外部交互响应解耦。该方案预测归一化各向异性刚度方向以进行定向柔顺性分配。此外，UniReflex集成了自适应门控机制，可实现以位置为主的规划与以力为主的执行之间的无缝过渡。真实世界双臂实验表明，UniReflex在保留原始位置精度的同时，显著提高了接触稳定性和成功率。在评估的主干网络上，与联合训练策略相比，我们的方法实现了每步反向延迟降低25至66倍。
英文摘要
Generative imitation learning policies excel at trajectory planning but lack closed-loop force regulation, while directly incorporating force modalities often requires redesigning or retraining the network. We present UniReflex, a universal plug-and-play framework that equips frozen generative policies with variable impedance control (VIC) for contact regulation, guided by force-direction intent collected during demonstration, without further slow-backbone fine-tuning. By non-invasively intercepting deep latent representations from the action head, UniReflex drives a fast reflex network that decouples active force exertion from external interaction response. This scheme predicts normalized anisotropic stiffness directions for directional compliance allocation. Furthermore, UniReflex integrates an adaptive gating mechanism that enables seamless transitions between position-dominant planning and force-dominant execution. Real-world bimanual experiments demonstrate that UniReflex significantly improves contact stability and success rates while preserving original position accuracy. Our approach achieves 25-66x lower per-step backward latency relative to joint training strategies on the evaluated backbones.
