抄録
An on-line motion planning algorithm is proposed for an autonomous mobile robot which allow it to approach its goal while avoiding multiple moving obstacles such as people and other mobile robots. The future trajectories of the obstacles depend on the future motion of the robot, since each obstacle also avoids other obstacles, including the robot. In other words, each obstacle has its own‘motion principle’which determines its motion from the location of its goal and the motion of other obstacles. In this study, the robot infers the principles of all obstacles in its sensing area from their trajectories observed in real time. This means it can predict their future trajectories, which are influenced by its future motion, using these principles. For a given motion principle and predicting the future trajectories of the obstacles, the robot evaluates the total time required to reach its goal and the degree of danger involved in avoiding obstacles. The robot tries this prediction and evaluation for several principles and then selects the optimum principle as the current one. Furthermore, it changes its principle at regular intervals in the above mentioned way. The difference between the real and inferred principles of the obstacle is reflected in the error of the predicted trajectory.
