A Path Planning Method for a Four-Wheeled Robot Based on an Intelligent Algorithm

An intelligent based local path planning algorithm is proposed enabling a four-wheeled differential mobile robot to avoid both static and moving obstacles. The developed intelligent method named as feedback compensation neural network (FCNN) refers to improving traditional back propagation neural network (BPNN). With respect to local path planning, the structure of FCNN is composed of two BPNNs. The first BPNN (main NN) is properly trained, playing a dominant role in dynamic obstacle avoidance. Subsequently, the second BPNN (compensation NN) is trained online in order to obtain the compensation for the output. Moreover, the FCNN can predict the motion of dynamic obstacles according to the previous obstacle avoidance situation, even if the obstacle velocity continuously changes. In such a way, the robot collision avoidance performance is highly enhanced, especially in the presence of moving obstacles changing their velocities and heading. The effectiveness of the presented algorithm is illustrated by simulation results.