Seeing Beyond the Present: A Future-Aware Privileged Knowledge Distillation Framework for Vehicle Trajectory Prediction
Yuxi Xue, Chao Wang, Xiaobo Chen, Jianjun Qian, Jian Jun Yang
Nanjing University of Science and Technology
阅读操作
确认中在文库中上传 PDF 后可生成中文音频讲解。
摘要与影响
Vehicle trajectory prediction serves as a core component in enhancing the safety of autonomous driving systems. Most existing methods focus on extracting historical context from trajectory data, while overlooking the importance of future context. To address this limitation, we propose a generic future-aware privileged knowledge distillation (FAPKD) framework to enhance existing trajectory prediction models that lack the capability of capturing potential future interactions. FAPKD includes a teacher, a student, and a knowledge distillation module. The teacher accesses true future states of neighboring vehicles as privileged information to capture real future context, while the student learns to infer it from past observations. The distillation module, with a multi-stage design and a distillation timestep selector (DTS), ensures that the student effectively learns to predict future context. Furthermore, to mitigate the prediction drift that often arises when applying Mean Squared Error (MSE)-based distillation to trajectory data, we introduce a wavelet-based trajectory distillation method. This method decomposes trajectory sequences into wavelet coefficients and subsequently represents them as probability distributions. By operating at the distribution level, the student is guided to internalize the teacher's latent knowledge while not deviating from the true future trajectory. Experimental results on the NGSIM, HighD and Argoverse 1 datasets demonstrate that our framework is model-agnostic and can be seamlessly integrated into existing encoder-decoder-based trajectory prediction models, leading to notable performance improvements.
逐年被引趋势
暂无年度引用数据
关键指标
同类平均 = 1
同领域 · 同年份 · 同类型
Google Scholar 与 OpenAlex 的被引统计范围不同,数值存在差异属正常。
AI 辅助阅读
依据:摘要
可就本文提问;依据不足时会说明。
学术脉络
学科主题
工程Autonomous Vehicle Technology and Safety
Traffic Prediction and Management Techniques · Traffic and Road Safety