In this age, when machine learning is so mature, it is a trend to use deep learning to work for people, not only to recognize various objects and things efficiently but also to apply in various fields and to reduce the burden for people. The YOLOV4 neural network was adopted to recognize in the first part of this research. Because of YOLOV4’s low computational power, no need to connect to a server, its ability to recognize objects accurately, and its fast computational characteristics, so it is suitable for this study. In order to promote the performance of obstacle avoidance. The second part of this research is to add the depth camera D435i which together with the YOLOV4 neural network, can more accurately measure the distance and depth relationship of obstacles ahead. The third part is to design a set of lightweight and smart controller to achieve the above mentioned obstacle recognition ability. The fourth part is the mechanical design of biped robot. Some simulation and experimental results are provided in this research. Furthermore, a biped robot prototype is implemented. We hope the implemented biped robot can not only replace human beings to complete high risk jobs, but also can assist in transporting materials.
