Shallow Neural Network Based Vision Mobile Robot Control

Zhe Wu ； Yiqin Yang ； Qingyang Xu ； Fabao Yan

《電腦學刊》 30卷4期 (2019/08) Pp. 78-89

https://doi.org/10.3966/199115992019083004007

摘要

Although deep neural network has many advantages, the main problem is the large requirement of computing power. The performance of shallow network is inferior to deep neural network, but the requirement of computing power is smaller. Therefore, we present an improved shallow neural network based on the deep learning techniques. The new techniques will improve the shallow neural network. This work takes a mobile robot as experimental platform using a camera module and a Raspberry Pi. The proposed shallow neural network is used to study driving a robot as a simulation research of unmanned vehicle, and then the robot can be steered by the trained neural network. In the training mode, the camera module provides the images needed for training the shallow neural network, and the robot will predict the movements and direction using the camera images according to the trained model in the autonomous mode. The experiment exhibits the effectiveness of the proposed model.

關鍵字

batch normalization ； mobile robot ； Raspberry Pi ； shallow neural network