1 paper
Guangxin Zhao, Yiran Hu, Yuan Cao +7
Existing LLM-driven robot task planners rely on a taken-for-granted assumption of an ideal user whose instructions are clear, complete, and task-focused. However, when interacting…