Development of autonomous features and indoor localization techniques for car-like mobile robots

Intelligent autonomous navigation in a large-scale and unknown indoor environment is an important problem in mobile robotics. For a car-like mobile robot with a racing car platform, the movement control concept is similar to that of a car.
The autonomous features enable robots to control own motion without human
interference. Three autonomous features are addressed in this thesis; obstacle avoidance, doing 180° turns in a narrow corridor, and path following control. Since the obstacle positions are not known beforehand, the strategy requires not only the obstacle avoidance but also trajectory generation and robot localization.
The robot localization can be broken down into relative and absolute localization. This thesis addresses the development of the model-based relative localization technique and the landmark-based absolute localization technique.
The model-based relative localization is applied by the non-linear dynamic car model to the Kalman filter. The study of integrating sensor data from odometer, gyroscope and compass for the position and heading estimators provides a discussion of the performance of three localization methods; differential drive, gyroscope estimator, and compass estimator.
The landmark-based absolute localization is applied by using the 3D camera and the 3D artificial landmark and is called the position calibration. Three parts of the position calibration are developed: The design of landmarks, the landmark recognition, and the robot position prediction and update. Lastly, the improvement for the resolution of the position calibration by using 2D and 3D images is studied.