Design and Implementation of an Autonomous Navigation System for Paquitop, a novel mobile robot for indoor assistance

The goal of this thesis is the development of an autonomous navigation system for a mobile robot named Paquitop. Its main peculiarity is to achieve omnidirectional motion exploiting conventional wheels. The Paquitop platform has been designed and prototyped by the researches at Politecnico di Torino during the last two years. This work represents the first step to enable the robot with autonomous navigation capability to perform indoor assistive tasks in full autonomy. The software architecture is designed to fulfill the following tasks: Mapping, Localization, Obstacle Avoidance and Path Planning. The first step for the design is the definition of a Hybrid Control Architecture based on the primitive functions Sense-Plan-Act. The basic idea behind this design is to decouple the navigation algorithms inside these primitive according to their task. Firstly, the platform extracts in real-time information perceived from the environment through the sensors, returning its local position estimate, the map of the surrounding environment and the obstacle detection. The ability for a mobile robot to localize itself and at the same time to build a map of the surrounding environment recognizing the obstacles, is defined Visual Simultaneous Localization and Mapping (VSLAM). A navigable path between the starting point and the target one is computed by the global planner. The global path is subdivided into suitable waypoints by the local planner, which takes into account the dynamic obstacles and the vehicle constraints. Finally, the planned actions are controlled until the goal is reached. The adopted framework for the implementation of all tasks is the Robot Operating System (ROS). Inside this framework, the Unified Robot Description Format (URDF) is used to represent the robot kinematic structure in the form of links and joints. The ROS packages used to fulfill the global task are included inside the Navigation Stack package. In conjunction with ROS, Gazebo simulator and RViz visualizer are adopted to simulate robot in its operational environment. To effectively achieve autonomous navigation, the Paquitop platform has been provided with exteroceptive sensors: a LiDAR is used to acquire data for the SLAM algorithm, a commercial tracking camera is adopted to estimate the pose changes in time through sensor fusion techniques and an on-board PC to provide the need computational capability. The brain of the whole system is the computer Nvidia Jetson Nano, connected to Arduino which receive and send information to electronic devices. The obtained result involves a both reactive and deliberative framework, which guarantees an accurate mapping and localization in the environment. This data can be used by the navigation algorithms to accomplish the mobile robot to the desired pose and to perform an obstacle avoidance logic. The thesis is divided as follows: the Chapter 1 presents the state of the art of mobile autonomous robots, the logic of the motion planning and the tools used to develop the thesis work; the Chapter 2 describes the Paquitop platform analyzing its kinematic and its main components; the Chapter 3 defines the software robotic architecture with its packages; the Chapter 4 describes the implementation of each package; the Chapter 5 focuses on validation tests; the Chapter 6 presents the conclusions pointing out the system limits and its possible improvements.