A new hybrid algorithm of dynamic obstacle avoidance based on dynamic rolling planning and RBFNN

Combining rolling planning principle with radial-basis-function-neural-network (RBFNN), this paper brings forward a new hybrid obstacle-avoidance algorithm, which aims at dynamic obstacles and can be utilized in local path planning for mobile robot in dynamic and uncertainty environment. The method uses camera to collect the motion path of dynamic obstacles, picks up heart-shaped sequence and then with these data, builds the prediction model using RBFNN. A rolling window in the scan range of ultrasonic sensor is built according to the present location in the case of the real-time plan of mobile robot. When the planner finds there is a dynamic obstacle in the rolling window, then begins the prediction calculation. It predicts the obstacle´s motion path in the next period based on three time sequence value of the obstacle in contiguous period of time. So the dynamic obstacle avoidance issue can be converted to the instantaneous static one and make the real-time plan feasible. The method can enhance the safety of dynamic avoidance and the real time of plan. The simulation results have proved the algorithm´s feasibility and efficiency.