Robust navigation for mobile robot during day and night

Abstract

This dissertation aims to provide a solution for robust navigation for mobile robot during day and night. The project consists of two main components - point clouds se mantic segmentation and imitation learning for velocity prediction. Features of point clouds and some related work of point clouds semantic segmentation are introduced in chapter 2. In chapter 3, we focus on the specific technologies including deep learn ing techniques and the structure of neural network we used in this project. For point clouds semantic segmentation part, we compare three similar network. And for veloc ity prediction part, we have tested MLP, LSTM and multi-head LSTM networks. An imaged based convolutional neural network is also used to make contrast experiment. DeepLab V3 which is used to automatically label the point clouds in this project is also introduced in this chapter. The results and analysis of all experiments are given in detail in chapter 4. Finally, chapter 5 makes a conclusion of all work done, presents current problems and possible solutions for future work.