Construct state-action map through human control trajectories and computation

Extracting, optimizing and transferring control skill is an important step towards creating an intelligent robot in a cooperative environment. Traditionally, control skill analysis focuses on how to design a state-action map to provide appropriate actions according to the states. Therefore approaches such as dynamic programming and neural networks are employed to compute the appropriate actions for all the states. However, this results in time consuming computation, low robustness and no human control characteristics. Although human control skill is robust, it is subcognitive and difficult to be described by the human operators themselves. In this case, human control trajectory is proposed to record and analyze human control skill. Human control trajectory is just a sequence of the systempsilas states and corresponding human operatorpsilas control actions, sampled at a certain frequency. Taking the advantages of human control trajectories and computation, a new approach is proposed. That is analyzing the human operatorspsila control trajectories to find out the most important states which limit the control effect. And then compute the selected states to set up the local state-action map. After that, based on the computational results and the human control trajectories, synthesize the global state-action map to realize the control skill transformation process. In this paper, a dynamic simulator is used as an example to evaluate the effect of the proposed approach. The results show that this approach can reduce the total computational cost while improve the human operatorpsilas control skill effectively.