Analytic Approach for Robot Control Using Natural Language in Dynamic Environment

Existing approaches of robot control by natural language assume a static environment which is not likely to happen when robots and human partners coexisting in a shared common area. Human behaviors may affect the robot execution and cause failure of current task plans. To improve adaptivity of natural language controlled systems to dynamic environments such that they can implement the assigned task under disturbances, we propose an analytic approach to model environment dynamics into high-level robot behavior model. The unforeseen events due to human activities can cause abrupt jump on task progress. The presented approach can regenerate an action sequence w.r.t. input natural language instructions and temporal task execution status. In addition, three properties: controllability, nonblockingness, and stability are used to evaluate and refine the behaviors of the robot controller. This guarantees the accessibility of the goal states corresponding to successfully accomplished task and the convergence to the goal states under dynamic environment. Experiments have been conducted to illustrate the proposed approach.