| Autonomous driving technology involves environmental perception,decision-making,trajectory planning,and control,and is a challenging issue in artificial intelligence.This paper mainly relies on the national key research and development program "Research and Practice of swarm intelligence Checked Vehicles for Airport Baggage Transfer"(2021YFE0203600),and studies the decision planning algorithm of intelligent vehicles in multiple driving scenarios.Modeling space dimension,vehicle system dynamics,interaction complexity of traffic flow environment and perceived uncertainty all lead to the complexity of automatic driving problem.Considering the complex vehicle dynamics model and all feasible actions of the nonlinear mobile robot system,it is difficult to generate motion planning strategies between arbitrary boundary states.Moreover,due to the objective existence of dynamic environments and perceptual errors,real-time trajectory planning is more challenging.Due to the fact that most autonomous driving products that have been implemented use decision modules such as finite state machines and decision trees,these often result in poor interpretability,maintainability,and scalability.This article adopts a layered architecture for decision-making and behavior planning in autonomous driving.By combining behavior blocks in a bottom-up manner into a more complex framework,decision algorithms based on specific scenarios and methods can be incorporated into the behavior blocks to achieve high coverage of autonomous driving decision scenarios.And the concept of decision makers is introduced to solve scenarios where multiple behavior block options are applicable but there is no clear priority between them.Then,a behavior generation stack was developed for autonomous driving in both structured and unstructured roads to guide autonomous vehicle decision-making in various scenarios.Trajectory planning is a key component of autonomous driving,directly responsible for driving safety and efficiency during driving.The ability to find the best track in real time is crucial to the auto drive system.Therefore,this paper will use fast iterative search and sampling strategies for sampling based track planning in urban and high-speed scenes,so that the best track can be found in real time and efficiently from a large number of candidate tracks.Before generating any trajectory,use historical planning results as prior information in the heuristic to estimate the cost distribution in the sampling space.On this basis,a fast iterative search and sampling strategy are adopted to explore possible candidate sampling spaces,and alternative trajectories are generated during the search process.For low-speed parking scenarios,this article adopts a dynamic search algorithm as the front-end coarse trajectory generation,and introduces the concept of spatiotemporal corridors to optimize the trajectory in a convex space in the back-end,thereby improving the quality and efficiency of the solution.For the downstream control module of the auto drive system,LQR(Linear quadratic form Regulator)is used to track the trajectory and verify the superiority of LQR compared with the pure tracking method.Finally,this article designs an overall simulation architecture based on Carla.The map uses OpenDRIVE map to search for global paths using a global planning algorithm adapted to high-precision maps,which is then used for subsequent verification of structured and unstructured road planning methods.Furthermore,due to the complexity of actual vehicle testing,this article first selects a small wire controlled chassis to communicate with software and hardware,and completes hardware in the loop experiments.The subsequent actual vehicle experiments will rely on the Yunle wire controlled chassis to test the algorithm capability of this article.By conducting planning and parking tests in the Weishui Campus of Chang’an University,and analyzing the results,the applicability and performance of the algorithm in this article will be evaluated. |
PDF Full Text Request