An Embedded Convolutional Neural Network for Maze Classification and Navigation

Traditionally, the maze solving robots employ ultrasonic sensors to detect walls around robot. The robot is able transverse along omnidirectionally measured depth. However, this approach only perceives presence of objects without recognizing type these objects. Therefore, computer vision has become more popular for classification purpose in applications. In study, a equipped with camera recognize types obstacles maze. are classified as: intersection, dead end, T junction, finish zone, start straight path, T–junction, left turn, and right turn. Convolutional neural network, consisting four convolution layers, three pooling fully-connected employed train using total 24,000 images obstacles. Jetson Nano development kit used implement trained model navigate results show an average training accuracy 82% time 30 minutes 15 seconds. As testing, lowest 90% T-junction computational being 500 milliseconds each frame. convolutional network adequate serve as classifier