Multi-Modal Architectures for Mobile Robot Localization Based on On-board Sensors

Nowadays, autonomous mobile robots (AMRs) are increasingly used in the industry to assist or even replace human operators during repetitive tasks. It is therefore essential that they can always be well-localized in the work-space in order to move reliably and safely. Providing accurate localization based on on-board sensors such as encoders, IMU, and LiDAR remains a challenge, as these sensors are affected by noise, bias, and drift, inherent in their structure and also caused by the surroundings. This thesis addresses the problem of AMR localization using different types of architectures, fully exploiting the available sensors and adapting them to the characteristics of the environment. The adopted approach considers odometry after it has been calibrated for the surface on which the robot moves, an extended Kalman filter, and a kinematic iterative closest-point localization algorithm, combined synergically to obtain a robust pose estimate through feedback systems. In particular, it results that a reliable odometry is an important component for the implementation of the state-of-the-art adaptive Monte Carlo localization algorithm, which serves as a reference for robot navigation and as a way to evaluate the robustness and accuracy of all localization estimates. In the thesis, all pose estimators are implemented in the ROS2 environment, through pre-configured packages, which ensure flexibility and adaptability to the project under consideration. Simulation tests are performed in Webots, an open-source 3D simulator, where the HELKO robot has been reproduced. Calibration tests were conducted at the Eureka System R&D laboratory, while comparative evaluations of the estimates were carried out at the SPARCS laboratory of the Department of Information Engineering.