Skip to main content
QUICK REVIEW

[论文解读] Stochastic Trajectory Prediction via Motion Indeterminacy Diffusion

Tianpei Gu, Guangyi Chen|arXiv (Cornell University)|Mar 25, 2022
Autonomous Vehicle Technology and Safety被引用 6
一句话总结

本文提出运动不确定性扩散(MID),一种新颖的随机轨迹预测框架,将人类运动的不确定性建模为反向扩散过程,从高斯噪声分布逐步去噪,生成多样化且高质量的轨迹。通过基于Transformer架构的条件马尔可夫链学习,MID在Stanford Drone和ETH/UCY基准上达到最先进性能,可通过调整扩散步长灵活控制预测的多样性与准确性。

ABSTRACT

Human behavior has the nature of indeterminacy, which requires the pedestrian trajectory prediction system to model the multi-modality of future motion states. Unlike existing stochastic trajectory prediction methods which usually use a latent variable to represent multi-modality, we explicitly simulate the process of human motion variation from indeterminate to determinate. In this paper, we present a new framework to formulate the trajectory prediction task as a reverse process of motion indeterminacy diffusion (MID), in which we progressively discard indeterminacy from all the walkable areas until reaching the desired trajectory. This process is learned with a parameterized Markov chain conditioned by the observed trajectories. We can adjust the length of the chain to control the degree of indeterminacy and balance the diversity and determinacy of the predictions. Specifically, we encode the history behavior information and the social interactions as a state embedding and devise a Transformer-based diffusion model to capture the temporal dependencies of trajectories. Extensive experiments on the human trajectory prediction benchmarks including the Stanford Drone and ETH/UCY datasets demonstrate the superiority of our method. Code is available at https://github.com/gutianpei/MID.

研究动机与目标

  • 为解决轨迹预测中建模多模态、不确定的人类运动的挑战。
  • 通过显式建模运动不确定性变化,克服现有方法(如GANs的训练不稳定性)和CVAEs(生成不切实际样本)的局限性。
  • 通过可控的扩散过程长度,实现预测多样性与确定性之间的灵活权衡。
  • 通过基于Transformer的架构,结合观测行为与社交互动,提升轨迹的时间建模能力。

提出的方法

  • 将轨迹预测建模为反向扩散过程,从可行走区域的噪声分布中逐步去除不确定性,最终得到确定性轨迹。
  • 使用参数化的马尔可夫链与高斯转移分布建模反向扩散过程,通过状态嵌入条件化于观测轨迹。
  • 利用时空图网络将历史轨迹与社交互动编码为状态嵌入。
  • 采用基于Transformer的主干网络捕捉轨迹序列中的长程时间依赖关系。
  • 使用变分推断与变分下界目标函数进行模型训练。
  • 在推理阶段通过从噪声分布逐步去噪,生成多样化且高质量的轨迹。
Figure 1 : Illustration of the reverse diffusion process for human motion indeterminacy variation. Under high indeterminacy, the trajectory distribution can be regarded as a noise Gaussian distribution which denotes ambiguous walkable areas. With the decreasing of indeterminacy, this distribution gr
Figure 1 : Illustration of the reverse diffusion process for human motion indeterminacy variation. Under high indeterminacy, the trajectory distribution can be regarded as a noise Gaussian distribution which denotes ambiguous walkable areas. With the decreasing of indeterminacy, this distribution gr

实验结果

研究问题

  • RQ1建模人类运动从高不确定性到低不确定性的演化过程,能否提升轨迹预测质量?
  • RQ2如何在随机轨迹预测中实现预测多样性与准确性的灵活权衡?
  • RQ3基于扩散的方法能否在轨迹预测中超越现有生成模型(如GANs和CVAEs)?
  • RQ4所提方法在多大程度上能够捕捉行人运动中的复杂时间与社交依赖关系?

主要发现

  • MID在Stanford Drone和ETH/UCY轨迹预测基准上达到最先进性能。
  • 通过调整反向扩散步数,方法可实现多样性与确定性的可控权衡。
  • 扩散链越长,轨迹多样性降低而预测确定性提高,符合预期行为。
  • 定性结果表明,MID生成的轨迹在长期预测中比Trajectron++更接近真实轨迹。
  • 模型生成的轨迹多样化且可行,在去噪后仍保持在可行走区域内。
  • 尽管推理时间较长(在ZARA1上生成512条轨迹耗时17.368秒),该方法仍能提供高质量、逼真的预测,适用于动态环境。
Figure 2 : The architecture of our MID framework. MID consists of a temporal-social encoder network and a Transformer-based decoder network. The encoder maps the history path and social interaction clues into a state embedding. The decoder then takes $\mathbf{y}_{k}$ along with state embedding and t
Figure 2 : The architecture of our MID framework. MID consists of a temporal-social encoder network and a Transformer-based decoder network. The encoder maps the history path and social interaction clues into a state embedding. The decoder then takes $\mathbf{y}_{k}$ along with state embedding and t

更好的研究,从现在开始

从阅读论文到最终审阅,大幅缩短您的研究时间。

无需绑定信用卡

本解读由 AI 生成,并经人工编辑审核。