Implementation Of Autonomous Navigation And Obstacle Avoidance On An Unmanned Ground Vehicle

Abstract

This thesis presents the implementation of a novel distributed embedded systems approach to real-time obstacle avoidance and guidance for an Unmanned Ground Vehicle (UGV). The mobility, real-time, and limited size requirements of UGVs, result in computationally limited and resource constrained hardware platform. The use of distributed computational resources, such as multiple embedded micro-controllers, enables the distribution of the computing resources for obstacle avoidance and guidance system functionalities. The resulting system's complexity is significantly greater than that of a single high performance processor performing all of the above functions. The hardware platform is integrated with sensors and micro-controllers to function as the real-time obstacle avoidance and guidance system for a UGV. The sensors include: a GPS receiver, a digital compass, rotary encoders, and a scanning laser range finder. All sensors have been calibrated and characterized for accuracy and reliability. The obstacle avoidance and guidance functionality executes on MPC555 micro-controller. The data strings from the sensors are parsed on IsoPod, PlugaPod micro-controllers. The required sensor data are passed over to the MPC555 over CAN network as part of a distributed computing architecture. A simulation model consisting of the guidance and navigation algorithm along with the tank model was developed. The simulation model performs obstacle avoidance and waypoint navigation successfully. A real-time model to perform obstacle avoidance and waypoint navigation was developed. The real-time model takes inputs as sensor data, constructs a dynamic map of the environment and outputs control signals to navigate the vehicle through obstacles and towards waypoints. The real-time system successfully performs waypoint navigation. The real-time systems constructs an inaccurate local map in real-time environment. An accurate local map is successfully constructed in simulation from the real world data. Due to the erroneous map constructed in real-time, the real-time system does not successfully navigate through the obstacles.