Developing a planning and a perception module for an autonomous warehouse mobile robot

This thesis aims to develop an autonomous warehouse mobile robot to help automate the assembly process at Vaderstad. There are two main contributions to this work. First, since the robot's onboard sensors have a limited field of view, factory ceilings are equipped with RGB cameras looking downward. A floor segmentation module takes the overhead RGB streams as input and segments the floor as free versus occluded pixels. In a subsequent step, the segmented images are fed into a floor perception module, which constructs the assembly line's global 2D cost map. With this global cost map, the robot can plan its path toward its goal without exploring. In the second part, we implement a path-planning algorithm for non-circular and holonomic robots. We achieve this through a motion primitive-based algorithm called Hybrid-A*. Finally, the entire system is rigorously tested in simulation and real-world applications for efficiency and compatibility with ROS2's navigation system