计算机工程与应用 ›› 2018, Vol. 54 ›› Issue (15): 131-138.DOI: 10.3778/j.issn.1002-8331.1704-0323

• 模式识别与人工智能 • 上一篇    下一篇

基于禁区惩罚函数和MPC倍增预测的车辆避障研究

花晓峰,段建民,田晓生   

  1. 北京工业大学 城市交通学院,北京 100124
  • 出版日期:2018-08-01 发布日期:2018-07-26

Research on vehicle obstacle avoidance based on restricted areas penalty function and MPC prediction multiplication

HUA Xiaofeng, DUAN Jianmin, TIAN Xiaosheng   

  1. College of Metropolitan Transportation, Beijing University of Technology, Beijing 100124, China
  • Online:2018-08-01 Published:2018-07-26

摘要: 为了增强无人驾驶汽车对障碍物预测及躲避的可靠性,以模型预测控制(MPC)和车辆动力学模型为基础,提出了一种基于障碍物禁区惩罚函数和MPC预测距离倍增法的避障控制策略。考虑车辆的尺寸,对于障碍物边界的确定引入以阶跃函数为基础的禁区边界惩罚函数。同时在障碍预测环节中对预测距离进行倍数扩增,提高了车辆在较远距离处对障碍物的预测能力。仿真实验以车辆动力学平台CarSim为基础,结合Matlab/Simulink的S函数,对不同仿真工况进行测试。实验结果表明车辆可以避开给定障碍物并能够返回到原始路径,结果达到预期要求,验证了算法的可行性。

关键词: 模型预测控制, 禁区边界, 倍增法, 避障

Abstract: To improve the reliability of predicting and avoiding obstacles of unmanned vehicle, based on MPC method and vehicle dynamic model, the paper puts forward an obstacle-avoidance-control strategy based on penalty function of restricted areas of the obstacles and multiplication of predictive distance. Taking the size of vehicle into consideration, the step function based penalty function of restricted boundaries is introduced for defining obstacle borders. Meanwhile, in the process of obstacle prediction, the predictive distance is multiplied, improving the ability of obstacle prediction of vehicle in a far distance. Simulation test is based on vehicle dynamics platform CarSim, combined with S-function in Matlab/Simulink. Different operating modes are tested. Results show the vehicle can avoid the obstacle given and return to the original path, which meets the expected requirement, verifying the feasibility of the algorithm.

Key words: Model Predictive Control(MPC), restricted boundaries, multiplication, obstacle avoidance