Enabling Autonomous Multi-Floor Navigation for Robots in ROS2 using Behavior Trees

In today's rapidly evolving world, service robots are assuming critical roles in diverse fields, ranging from manufacturing, agriculture to healthcare. These mobile robots are required to process advanced navigation capabilities, allowing them to move seamlessly within multi-floor layouts. While traditional navigation solutions have excellent development in guiding robots through single-level environments, the challenge lies in enabling autonomous robots to understand the structural elements connecting different floors, such as elevators and stairs. The research work addressing these problems has been developed in collaboration with the LINKS Foundation, that conducts exploration across various domains, such as AI, IoT, Cyber-physical and autonomous systems. This thesis aims to investigate existing navigation approaches and explore novel solutions to recognize and interact with the environment (e.g., with automated doors, furniture, and elevators) to enable navigation across floors in a building using autonomous ground vehicle(AGV). The Robot Operating System 2 (ROS2) Navigation2 stack serves as a powerful tool for robot's navigation, offering a wide range of functions and algorithms for path planing and obstacle avoidance. Additionally, behavior trees have emerged as a high-level control framework for designing and modelling the robot actions, enabling the implementation of complex behaviors while providing insight into the execution process. This thesis extends the existing open-source ROS2 libraries by incorporating behavior-tree based behaviors and navigation planners/controllers to support a continuous navigation across floors. Simulation has been tested in Gazebo and Rviz with the two-storey office building at LINKS in Turin chosen as the environment. Following that, the application of these concepts to real-world robot navigation and a discussion on their applicability are included.