State Estimation Using the CoG Candidates for Sit-to-Stand Support System User

摘要

Various support systems have been developed to support elderly people, and the demand for indoor support system has increased. It is important to support not only walking but also to support sit-to-stand and stand-to-sit motions. We develop a support system for indoor use that depends on the user's state, such as sitting or standing. Although it is useful for assistive devices to be able to select how to support users based on sensor data, it is difficult to utilize many expensive and sophisticated sensors for accurate estimation of the user's state. In this study, we propose an estimation method of the user's state utilizing a few inexpensive and simple sensors. First, we propose the method to calculate the CoG candidates using a human link model. The CoG candidates are then used to develop a state estimation method for sit-to-stand motion; this motion consists of three contiguous states: sitting, rising, and standing. A support vector machine is used to estimate the user state and the methods were experimentally validated using the developed assistive robot. The experimental results show that the estimations are correct except in the vicinities of state transitions. The average state transition time errors are 0.175 and 0.145 s for sit-to-rise and rise-to-stand, respectively. Since sit-to-stand motion is contiguous, the user's state is ambiguous and can be both states at the boundaries. Therefore, the accuracy of the state estimation is reasonable.