Construction of a mind model for human-robot cooperation

Abstract

Directing person's behavior is one of the roles of mind which is dominated by emotion. In order that a robot assists a person cooperatively, it is necessary to predict what behavior the person is going to do. Therefore, the purpose of this study is to construct a mind model in order to identify person's mind for predicting or estimating person's behavior. In a process of behavior decision from the planning to the performing, it is considered that the role of mind is modifying original desire (Desired Input) of rather selfish action according to restriction (Restrictive Input) from the external world viewed from the brain, and issuing cooperatively a reference order of action which should be taken. The external world is divided into three sub-worlds like "subject's own body", "partner" and "concerned environment" in this research, and these sub-worlds are thought to be common situations of general cooperative work. Proposed mind model, hence, is designed to consist of three-layers corresponding to each external sub-world. Each layer of mind tries to take a balance between Desired Input and Restrictive Inputs. The role of taking the balance at each layer is expressed as an optimal control, which minimizes a performance index of quadratic integral form with the weight named "Emotion Index." To confirm effectiveness of this mind model in a cooperative work between a person and a robot, the proposed mind model is installed in an exoskeleton power assistive system named HAL (Hybrid Assistive Leg) for walking aid. Since HAL is firmly worn on operator's legs, HAL needs a cooperative control to play like a part of operator's body. In order to realize the cooperative power assist, it is useful to identify mind parameters which affect operator's behavior. Experiments of identifying the Emotion Index and the estimating Desired Input of third layer were carried out by using information from sensors built in HAL. And results of experiments show that the constructed mind model has given HAL the ability to identify and to estimate parameters of operator's mind.