基于CUDA并行的智能车辆MPC轨迹规划算法  

Intelligent Vehicle MPC Trajectory Planning Algorithm Based on the Parallel CUDA

作　　者：周妍 梁华为[1] 赵盼[1] 李碧春[1] 余彪[1] ZHOU Yan;LIANG Hua-wei;ZHAO Pan;LI Bi-chun;YU Biao(Institute of Applied Technology,Hefei Institutes of Physical Science,Chinese Academy of Sciences,Hefei 230088,China;University of Science and Technology of China,Hefei 230026,China)

机构地区：[1]中国科学院合肥物质科学研究院应用技术研究所,安徽合肥230088 [2]中国科学技术大学,安徽合肥230026

出　　处：《仪表技术》2019年第10期20-24,共5页Instrumentation Technology

基　　金：公安部消防局项目(2017XFGG04);国家重点研发计划(2016YFD0701401,2017YFD0700303,2018YFD0700602);中科院青促会项目(2017488);中科院135项目(KP-2019-16);安徽省新能源汽车暨智能网联汽车产业技术创新工程项目

摘　　要：模型预测控制(Model Predictive Cmitrol-MPC)轨迹规划算法涉及复杂的优化过程,易导致过多的计算负担,同时随采样密集度增大规划耗时成倍增长。为了提升轨迹规划效率,在满足规划实时性的前提下尽可能采样更加密集的轨迹簇以改善最终的规划结果,提出了基于CUDA并行的智能车辆MPC轨迹规划算法,在CUDA架构中实现轨迹生成和代价评估的并行设计,代价评估筛选与障碍物不相碰撞的平滑轨迹,确保得到的最优轨迹可行可靠。测试表明,该算法得到的规划结果是可靠的,且对比算法的CPU实现加速比提升了8倍。The trajectory planning algorithm based on model predictive control (MPC) involves a complicated opti-mization process, which is easy to cause excessive computational burden, and the planning time increases exponentially with the increase of sampling density. In order to improve the efficiency of trajectory planning, the denser trajectory clusters are sampled as much as possible to improve the final planning results. The parallel CUDA based intelligent vehi-cle MPC trajectory planning algorithm is proposed to realize trajectory generation in CUDA architecture and the parallel design of the cost assessment. The cost assessment screens the smooth trajectory that does not collide with the obstacles, thus it ensures the optimal trajectory obtained is feasible and reliable. Tests show that the proposed algorithm is reliable, and the CPU speed of the comparison algorithm is improved by 8 times.

关 键 词：轨迹规划 模型预测控制 CUDA并行 智能车辆 

分 类 号：U463.6[机械工程—车辆工程]