A Natural Semantic Task Understanding and Scheduling System for Crowd Robots

With the rapid development of robotic technology and the increasing demand for intelligent services, crowd robots have emerged as a promising solution for multi-robot cooperation. This study introduces a novel approach that combines large language models and reinforcement learning for natural semantic task understanding and scheduling in crowd-robot systems. Our system effectively interprets complex and ambiguous tasks described in natural language, facilitating task allocation to robots in various scenarios by adjusting the prompt template. With reinforcement learning, the system optimizes robot resource utilization and user experience from a global and long-term perspective. Our system combines robust security and reliability with a convenient user interface. Experiments on real-world datasets demonstrate that our system could accurately understand natural semantic tasks and efficiently schedule crowd robots. Case study and System Usability Scale (SUS) surveys illustrate the effectiveness and usability of our system, revealing its significant practical value in accurately parsing task commands and markedly improving task execution efficiency. Moreover, this research has considerable implications for the further development and application of crowd robots.