汽车工程 ›› 2024, Vol. 46 ›› Issue (6): 965-974.doi: 10.19562/j.chinasae.qcgc.2024.06.003

• • 上一篇    下一篇

基于多模态轨迹预测的智能车轨迹规划研究

黄晶(),刘祥臻,邓潇阳,陈然   

  1. 湖南大学机械与运载工程学院,长沙 410082
  • 收稿日期:2023-12-05 修回日期:2024-02-25 出版日期:2024-06-25 发布日期:2024-06-19
  • 通讯作者: 黄晶 E-mail:huangjing926@hnu.edu.cn
  • 基金资助:
    国家自然科学基金(52175088);国家杰出青年科学基金(52325211)

Research on Intelligent Vehicle Trajectory Planning Based on Multimodal Trajectory Prediction

Jing Huang(),Xiangzhen Liu,Xiaoyang Deng,Ran Chen   

  1. College of Mechanical and Vehicle Engineering,Hunan University,Changsha 410082
  • Received:2023-12-05 Revised:2024-02-25 Online:2024-06-25 Published:2024-06-19
  • Contact: Jing Huang E-mail:huangjing926@hnu.edu.cn

摘要:

混合交通流下由于驾驶员意图的不确定性行驶轨迹将呈现多模态属性,为了提高安全性并实现个性化驾驶,本文提出一种基于环境车辆多模态轨迹预测的智能车轨迹规划算法。首先,结合图卷积神经网络(GCN)和长短期记忆网络(LSTM)并加入注意力机制建立轨迹预测模型,预测不同行驶意图下的未来轨迹概率分布。然后,针对环境车辆的多意图概率下预测轨迹集合,根据自动驾驶风格偏好,设定一定的概率阈值挑选出确信轨迹,将其投影到规划路径上生成S-T图,并通过动态规划和二次规划进行基于碰撞风险规避的速度规划。最后,基于模型预测控制(MPC)对本文模型在典型换道场景和NGSIM真实道路场景下进行仿真测试并与现有模型进行对比验证。结果表明:本文提出的模型在安全性、舒适性和行车效率等方面均优于对比模型,能够在准确预测环境车辆未来轨迹的前提下实现最优轨迹规划,保证自动驾驶汽车安全、高效的行驶。

关键词: 自动驾驶, 轨迹规划, 多模态轨迹预测, 驾驶意图

Abstract:

Due to the uncertainty of the driver's intention under the mixed traffic flow, the driving trajectory will present multimodal attributes. In order to improve safety and realize personalized driving, a trajectory planning algorithm for intelligent vehicle based on the multimodal trajectory prediction of environmental vehicles is proposed in this paper. Firstly, a trajectory prediction model is established by combining graph convolutional neural network (GCN) and long short-term memory network (LSTM) with attention mechanism to predict the probability distribution of future trajectories under different types of driving intention. Then, for the set of predicted trajectories under multi-intention probabilities of environmental vehicles, a certain probability threshold is set to select sure trajectories according to the automatic driving style preference, which is projected onto the planning path to generate the S-T diagram, and speed planning based on collision risk avoidance is carried out through dynamic planning and quadratic planning. Finally, based on the model predictive control (MPC), the model proposed in this paper is simulated and tested in typical lane changing scenarios and real-road scenarios of NGSIM and compared with the existing model for validation. The results show that the model proposed in this paper is better than the model in comparison in terms of safety, comfort, and driving efficiency, which can realize the optimal trajectory planning under the premise of accurately predicting future trajectories of the environmental vehicles to ensure safe and efficient driving of autonomous vehicles.

Key words: autonomous driving, trajectory planning, multimodal trajectory prediction, driving intention