Research Of Trajectory Planning For Autonomous Driving Based Frenet Coordinate And Sampling

Trajectory planning is an important part of autonomous driving decision modules.How to generate an efficient,smooth and reasonable planning trajectory is one of the key technologies to achieve high level autonomous driving.In this thesis,Frenet coordinate system is integrated and a fifth-order differential equation is applied to solve the motion planning equation.The discrete motion states of the autonomous vehicle are available with sampling method.Furthermore,four typical scenarios in autonomous driving which include adaptive cruise control,lane change,stop and fixed velocity cruise are analyzed and visualized with proposed motion planning algorithm.The main research achievement of this thesis can be concluded as:(1)The coordinate transformation is deduced strictly between global Cartesian system and Frenet coordinate system.To achieve the trajectory planning,the kinematic attributes(such as the position,velocity and acceleration of autonomous vehicle)are transformed into Frenet coordinate.With Frenet representation,the state of vehicles can be easily differentiable and modeled.(2)The motion planning model of autonomous vehicles is made in both lateral and longitudinal directions based on state sampling strategy.After solving motion planning equations in lateral and longitudinal,the trajectories can be available.Furthermore,the trajectory quality evaluation models in lateral and longitudinal direction are developed to generate the optimal planned trajectories.(3)The steps of synthesizing the lateral and longitudinal planned trajectories are detailed under various scenarios.In order to predict the environment vehicles’ future trajectories,Kalman filter and recurrent neural network method are used.In addition,a three-circle method is proposed to judge whether the self-driving vehicle collides with the environmental vehicles,which is low cost and high efficient.(4)A trajectory planning algorithm simulation environment is set up to construct a three-lane road map,which is up to eight kilometers.Users can set geometry parameters and related vehicle dynamic parameters.Under the four typical scenarios,the parameters are configured,such as time,position,velocity and acceleration of the environmental vehicle and the optimal trajectory calculation and visualization analysis of the trajectory planning is performed.Simulation process is definitely robust and implemented in close-loop.Experimental results show that trajectories generated by the motion planning algorithm proposed in this thesis is stable,safe and executable.Compared with the third-order polynomial methods,the fifth-order polynomial motion planning algorithm has higher order control,and the generated trajectory is superior in quality and smoothness.The fifth-order polynomial motion planning algorithm is better than the seventh-order polynomial which is 16 Hz,increased the running frequency with 68.75%.The trajectory planning model of autonomous vehicles can simplify in Frenet coordinate.It is will be widely applied to other intelligent field.