Trajectory planning for a self-driving Electrical Vehicle: Design and development of a trajectory planning algorithm starting from the Occupancy Grid Map

During the 21st century, a consistent technological breakout has allowed a meaningful development of autonomous driving systems; this is due to state-of-the-art technologies which have contributed to the mentioned technological progress: the evolution of advanced sensors, definition of increasingly sophisticated technologies and an increase of computational power in generals influenced autonomous driving research field to reach advanced and challenging purposes. Self-driving systems could radically transform the current idea of transportation system, which inevitably would imply a significant change of our economy and society. Level 4 self-driving cars, which according to some automaker companies estimates, may be placed on the market in the next several years, would cause a complete social, economic, and technological revolution. With this thesis project, we want to cover a specific problem of the huge world hiding behind the autonomous driving. By presenting a specific designed algorithm, the trajectory planning field for autonomous driving systems is addressed. At high level, an autonomous driving system may be described by the so-called sense-act-plan procedure. The “sense” part is related to the sensors management, so that the vehicle surrounding environment can be detected the most effective way; the “act” part consists in the actuation system management; the “plan” part is where this work can be collocated. The project (VEGA: standing for “VEicoli a Guida Autonoma”) has been developed in Bylogix srl, a company which provides electrical and electronic engineering services and solutions for the Automotive industry, with a specific focus on autonomous driving. The core of this work is the design of a trajectory planning algorithm, once data from the sensors are received and processed. In particular, by means of a LiDAR the outdoor environment, including obstacles, is sensed and an occupancy grid map is given as input to the trajectory planner. Occupancy grid data, updated with a frequency of 10 HZ, are processed; thus, every 100 ms a trajectory is generated from the current vehicle position to a reference waypoint. From the generated trajectory, which guarantees the obstacles avoidance, the actuation variables are returned and sent by means of a CAN bus to the vehicle Electronic Control Unit. After an overview related to the experimental setup (hardware devices and software technologies used in the implementation phase) and after a detailed description of the trajectory planning algorithm implementation, the final obtained result is presented. The prototype vehicle acquired the capability of driving, fully autonomously, in a predefined path generating a collision-free trajectory, avoiding the obstacles on the path. The reference output variables, that are steering angle and vehicle speed, are shown compared with the actual measured values: such comparison has proven to be more than satisfactory.