Stochastic prediction of human motion and control of robots in the service of human

The authors propose a control model for human cooperative robots. According to human motion which is measured by a human recognition system, future human position is predicted. Robot trajectories are changed by acceleration/deceleration/stop so as to minimize danger, which is computed by the predicted data. In this paper, a model of a stochastic process is adopted for the prediction because it can prevent inappropriate robot motion which might be caused by prediction error. A prediction plane is divided into square cells. Each cell has probability data by which cell number, direction and speed change. They are determined by the characteristics of motion of places beforehand. Computation of the stochastic state transition chain predicts the probability that a human exists in a cell at future time. Application to a laboratory room demonstrated that this method is effective for safe motion of human cooperative robots