Control of human cooperative robots based on stochastic prediction of human motion

The authors propose a control model for human cooperative robots. In this model, the future human position is predicted on the basis of the measured human motion by a human recognition system. Robot trajectories are modified to improve safety which is computed using the prediction result. In this paper, a prediction method of stochastic process is adopted for the control model. In a room which is divided into square cells, a human state variable (cell number, direction and speed of motion) is stochastically made transitions as a Markov process. Simulation was performed for a room where a man and a robot are working together. The results demonstrated that the stochastic prediction is very effective for planning robot trajectories against danger, by which the robot can predict danger much earlier than by using the deterministic prediction method