Autonomous unmanned vehicle in indoor environment

The autonomous vehicle or unmanned ground vehicle more specifically to be implemented should be able to map an unknown environment and later navigate from an input source location to a destination location with obstacle avoidance ability. SLAM (Simultaneous Localization and Mapping) is the core technique adopted, which builds a map of its working environment and then localize itself in it. It estimates odometry by tracking features in the environment and increments the map simultaneously. By detecting a loop closure, previously visited places can be used to reduce map errors. With a mix of different proprioceptive, the mapping estimation would be more robust.[1][2] However, due to cost consideration, only one RGB-D camera was applied in this project. Visual odometry estimation was used at the cost of lost odometry. Since the sensor chosen was one RGB-D camera, it contained depth information of the surrounding as well. The depth information could be aligned with the color frames such that each pixel of the color frame would have a corresponding depth value. Taking advantage of this feature, a simple obstacle avoidance algorithm was introduced. Once the map of the environment was constructed and saved, the coordinate of the destination point could be set based on it. By comparing the current view with the map previously generated, the UGV could realize its current location inside the map. A routing strategy would then be introduced to navigate the rover with the avoidance algorithm previously introduced.Bachelor of Engineering (Electrical and Electronic Engineering