Trajectory generation for safe overtaking maneuver inautonomous vehicles : Evaluated in lane merging scenario utilizing a trajectory planner

More than 1.2 million people die each year due to road traffic injuries [1]. In order to reduce traffic accidents and human errors, autonomous vehicles is been the subject of intense research. To improve the driving experience, automotive companies have developed Advanced Driver Assistance Systems (ADAS) such as Adaptive Cruise Control (ACC) and Lane Keeping Aid (LKA) which aim to make driving safer and more comfortable. One particularly interesting maneuver is the lane change. Lane change maneuver is one of the riskiest maneuvers that a driver has to perform on a highway, and can be perceived as challenging since it involves changes in both the longitudinal and lateral velocities, direction and as well as movement in the presence of other moving vehicles. This thesis seeks to evaluate how different prediction model of the trajectory planner will affect collision risk, comfort and result in an increasing rate of successful overtakes. An trajectory-planning algorithm will be reliable in making smarter decisions for performing a safe overtaking maneuver’s and constantly generate discrete trajectory profile with respect to the parameters of the vehicle in front. Future motion is predicted using prediction models linking control inputs, vehicle properties and external conditions to the evolution of the state of the vehicle. The vehicle should be able to avoid collisions at the point of convergence where two lane road merges into a single lane road and therefore, motion only in the longitudinal direction is considered for the evaluation. The prediction model chosen for this thesis is constant acceleration (CA) and constant velocity (CV). The project is part of a large EU-project called SafeCOP (Safe Cooperating CyberPhysical Systems) usingWireless Communication which aims at developing a complete prototype of an intelligent transport system. A great amount of trajectory generation techniques have been surveyed and quartic polynomial is selected for trajectory generation as it has many benefits of having a low computational cost and the continuous concatenation of curves is possible. It is important in the trajectory planner to cancel out trajectories which would dynamically not be feasible and result in an increase risk of collision with the surrounding vehicle. The two chosen prediction models were evaluated for three different scenarios on which they are tested and their results is compared. For the different scenarios addressed in this thesis Constant Acceleration (CA) prediction model gave better result when compared to Constant Velocity (CV) prediction model and had an lower risk of collision which increases the number of successful overtakes. While doing so the jerk dynamic constraints were always considered to ensure that the trajectory generated are within the comfort zone of the passenger