Small obstacle size prediction based on a GA-BP neural network.

Accurate and effective acquisition of obstacle size parameters is the basis for environment perception, path planning, and autonomous navigation of mobile robots, and is the key to improve the walking performance of mobile robots. In this paper, a generic algorithm-back propagation (GA-BP) neural network-based method for small obstacle size prediction is proposed for mobile robots to perceive the environment quantitatively. A machine vision-based small obstacle size measurement experiment was designed, and 228 sets of sample data were obtained. A genetic algorithm optimized back propagation neural network was used to build a small obstacle size prediction model with obstacle pixel width, pixel height, pixel area, and obstacle-to-camera distance as input parameters and actual obstacle width, actual height, and actual area as output parameters. The results show that the correlation coefficient () between the predicted and expected values of the test data is higher than 0.999, the root mean square error is lower than 5.573, and the mean absolute percentage error is lower than 2.84%. The good agreement between its predicted and expected values indicates that the model can accurately predict the size of small obstacles. The GA-BP neural network-based small obstacle size prediction method proposed in this paper is simple to execute, has good real-time performance, and provides a new, to the best of our knowledge, way of thinking for mobile robots to acquire environmental data.