人間の障害物回避軌道の調査に基づいたファジィ推論による移動ロボットの障害物回避

抄録

To coexist with human, a robot has to avoid obstacles based on human-like flexible decision-making. In this article, human trajectories to avoid an obstacle were observed from above by a video camera. The resultant trajectories were finely smooth. Using those, fuzzy rules to decide the moving direction at every moment were derived as follows: as the input variables, the predicted closest distance (DCPA: Distance to Closest Point of Approach) of goal (x1), that of obstacle (x3), the elapsed time to the predicted closest point (TCPA: Time to Closest Point of Approach) of goal (x2), and that of obstacle (x4), are adopted. As the output variable, steering angle of robot (y) is adopted. Based on fuzzy-neural networks method, a network of 4 inputs (x1～x4) and 1 output (y) is prepared. A membership function of each variable has 5 isosceles triangles. Fuzzy rules, number of which is 625 (=5⁴), are assumed. Optimal center and width of each triangle are obtained so as that the network reproduces the experimental human trajectories with minimum errors. It was proven that the obtained fuzzy rules with tuned membership functions successfully produce the trajectory which is similar to human ones, which is better than the trajectory obtained by potential method in viewpoint of smoothness.