Examination Committee

Prof Ming LIU, ECE/HKUST (Chairperson)
Prof Zexiang LI, ECE/HKUST (Thesis Supervisor)
Prof Shaojie SHEN, ECE/HKUST (Thesis Co-supervisor)
Prof Fu ZHANG, ECE/HKUST
 

Abstract

This thesis presents a visual navigation system for service robots offering guiding services within indoor environments like banks. Such system is constituted by an obstacle map server, a localization module and a command generation module. For commercial deployment with reasonable cost, these modules are implemented on an upward-facing camera for visual localization, a forward-facing RGB-D camera for obstacle avoidance as well as two encoders for odometry. 
 
To estimate the location of robot, odometry information and visual measurements are constantly fused via an Extended Kalman Filter(EKF). The global consistency is highlighted since a simple EKF will gradually drift due to accumulation of estimation error. To this end, two complementary solutions are proposed 1) Quick Response(QR) code markers-based localization; 2) natural features-based localization. Though the first solution is much more robust to the changes of perceptual conditions like illumination and view angles, the second serves as a backup plan when pre-set markers are not available or strictly forbidden.
 
The obstacle map is offline built and online updated according to observations from the RGB-D sensor, allowing the command generation module to figure out a globally optimal path to goal location through classical A* algorithm, followed by a Dynamic Window Approach(DWA) to generate local commands to execute the path. 
 
According to experiments conducted in a few spaces where there exist movable chairs, tables, and pedestrians, the proposed system demonstrates real-time efficiency and low repeat errors(less than 10cm), as well as capability of avoiding dynamic objects. It is therefore concluded that the proposed system is a complete solution for the robot to offer long-term guiding services in dynamic indoor environment.