Revisiting Open-Loop Execution in Robotics: Toward Reactive, Higher-Performing Policies
技术动态
来源：arXiv cs.RO发布时间待核实
arXiv:2608.15938v2 Announce Type: replace Abstract: Action chunking --- the practice of predicting a sequence of actions and executing a prefix open-loop --- has emerged as a key enabler of recent progress in imitation learning for robotic manipulation. However, executing long open-loop prefixes reduces reactivity, limiting policies' ability to correct for errors.