Agent navigation is used in various artificial intelligence applications and research projects, for example, unmanned aerial vehicles, autonomous cars. Navigation as a whole involves multiple goals, for example, target reaching or task accomplishment, efficient path planning and collision avoidance. In terms of collision avoidance in dynamic environments, intensive calculation is needed using a computer vision based algorithm.

Problem Statement

In a dynamic environment, an agent should make fast decision of movement according to visual input. Therefore we will use deep learning and produce an end-to-end decision making neural network that does not depend on or use minimal object recognition in computer vision.


This project particularly serves as one component of Simultaneous Localization and Mapping (SLAM) infrastructure). We will focus on tuning the neural network for better performance in indoor environment and assume there is only one agent that is using such network in the environment.

About Us

We are two final year CS students, Cao Chao and Hu Shengjie, from the University of Hong Kong (HKU) and we are doing our final year project under the supervision of Prof. Wenping Wang (@WenPing Wang)from the Computer Science Department at HKU.

Please contact us at:
Cao Chao:
Hu Shengjie: