摘要
针对目前基于监督学习的轨迹预测模型数据利用效率低、精度有限的问题,提出一种轨迹预测模型及通用的自监督预训练策略。首先,基于Transformer搭建轻量化的轨迹预测模型,实现场景时序空间特征提取与交互关系建模;其次,设计运动信息时序掩码、道路信息空间掩码、交互关系掩码3类掩码重建任务对模型进行自监督预训练,以提升模型对场景通用特征的提取能力;最后,以预训练权重为初始化参数在下游任务中进行监督学习微调。在Argoverse2 Motion Forecasting数据集的实验表明,模型在预训练任务中能够很好地重建出交通场景,引入自监督预训练能够有效提升预测精度和数据利用效率,且对不同预测任务具有通用性,在单目标轨迹预测与多目标轨迹预测任务上minFDE6指标分别提升3.3%与3.7%。
To address limitation in prediction accuracy and data utilization efficiency of supervised learn-ing-based trajectory prediction models,a trajectory prediction model and a general self-supervised pretraining strate-gy are proposed.Firstly,a lightweight trajectory prediction model based on Transformer is established to extract tem-poral-spatial features while modeling interaction relationship.Secondly,three types of masks,namely motion infor-mation temporal mask,road information spatial mask,and interaction relationship mask,are designed for self-su-pervised pre-training tasks on the model to enhance the model's ability to extract general scene features.Finally,pretraining weights are used as initialization parameters for supervised learning fine-tuning in downstream tasks.Ex-perimental results on the Argoverse2 Motion Forecasting dataset show that the model can effectively reconstruct traf-fic scenes in pretraining tasks.The introduction of self-supervised pretraining improves prediction accuracy and data utilization efficiency.Moreover,it exhibits universality for different prediction tasks,achieving a 3.3%and 3.7%improvement in the minFDE6 for single-agent and multi-agent trajectory prediction tasks,respectively.
作者
李琳辉
付一帆
王霆
王雪成
连静
Li Linhui;Fu Yifan;Wang Ting;Wang Xuecheng;Lian Jing(School of Mechanical Engineering,Dalian University of Technology,Dalian 116024)
出处
《汽车工程》
EI
CSCD
北大核心
2024年第7期1219-1227,共9页
Automotive Engineering
基金
国家自然科学基金(52172382)
中央高校基本科研业务费项目(DUT22JC09)
辽宁省科学技术计划项目(2022JH1/10400030)资助。
关键词
自动驾驶
轨迹预测
自监督预训练
autonomous driving
trajectory prediction
self-supervised pretraining
