Research On Planning Algorithms For Autonomous Vehicles In Specific Urban Scenarios

The planning algorithm is a very important part of the autonomous driving system.The planning algorithm’s ability to handle complex scenarios determines the intelligence of the autonomous vehicle.Research on planning algorithms is a hot and difficult point in the research of autonomous driving.The development of autonomous driving technology depends on the continuous improvement of the development of planning algorithms.This article relies on the national key research and development plan of the Ministry of Science and Technology,"Research and Demonstration of Key Technologies of Electric Autonomous Vehicles" to study autonomous vehicle planning algorithms.Aiming at the problems that are not good enough for the variant scenario and the optimization of the planning trajectory in the current planning algorithms,the hierarchical state machine and model predictive control are applied to the planning algorithm to establish a set of reasonable and safe outputs that can respond to specific city scenarios.The planning algorithm framework makes the planning algorithm have good scalability for different scenarios,a high degree of optimization for solving trajectories and higher comfort level.First,this paper chooses a hierarchical state machine as the method of behavior planning.The top layer accomplishes state division based on high-precision map information,determines the state transform conditions in the high-precision map,and illustrates the construction method of some high-precision map information.The middle layer uses Rule-based state division and state transform.According to the state of the top layer,the state of the middle layer and state transform conditions are divided in a rule-based manner,and the planning tasks within the state are determined.The hierarchical state machine is implemented by C ++ language,and then the advantages of extensibility of hierarchical state machine based on scene division are discussed.Then,by applying the responsibility-sensitive Safety model,the results of behavior planning are converted into the reference trajectory as the input of motion planning.Then referring to the method of dimensionality reduction,the three-dimensional optimization problem of motion planning is converted into two two-dimensional optimization problems,and the three-dimensional optimization problem of S-L-t is transformed into the S-t and S-L problem.Solve and define the two problems as model predictive control problems.Then select the predictive model and combine the reference trajectory of the behavioral planning to get the solution of the motion planning and the expected trajectory in the planning time domain.Later,using ROS、Prescan and Simulink joint simulation,by selecting specific urban scenarios,the rationality,safety,and comfort of the trajectories output by the researched planning algorithm when facing different scenarios were tested and evaluated.The effectiveness of the proposed planning algorithm is verified by simulation experiments.Then,a real vehicle test platform is built to verify the performance of the planning algorithm in the presence of control errors.The comprehensive analysis validates the effectiveness of the proposed hierarchical state machine plus optimization method as a planning algorithm.