Predicting the future behavior of human road users remains an open challenge for the development of risk-aware autonomous vehicles. An important aspect of this challenge is effectively capturing the uncertainty inherent to human behavior. This paper proposes an approach for probabilistic trajectory prediction based on normalizing flows, which provides an analytical expression of the learned distribution. We reformulate the problem of capturing distributions over trajectories into capturing distributions over abstracted trajectory features using an autoencoder, simplifying the learning task of the normalizing flows. TrajFlow improves the calibration of the learned distributions while achieving predictive performance on par with or superior to state-of-the-art methods on the ETH/UCY and the rounD data set.

翻译：预测人类道路使用者的未来行为仍是发展风险感知自动驾驶车辆面临的开放挑战。该挑战的一个重要方面是有效捕捉人类行为固有的不确定性。本文提出了一种基于归一化流的概率轨迹预测方法，该方法能够提供所学分布的解析表达式。我们将捕捉轨迹分布的问题重新表述为利用自编码器捕捉抽象轨迹特征的分布，从而简化了归一化流的学习任务。TrajFlow在提高所学分布校准性的同时，在ETH/UCY和rounD数据集上实现了与最先进方法相当或更优的预测性能。