vix.ing · top · new · best · stats · spec

Prediction by Anticipation: An Action-Conditional Prediction Method\n based on Interaction Learning

2020/12/24 by Ershad Banijamali, Banijamali, Ershad, Mohsen Rohani +7
Computer Science · Engineering · #Anomaly Detection Techniques and Applications #Autonomous Vehicle Technology and Safety #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Human Pose and Action Recognition #Machine Learning (cs.LG) #Online Learning and Analytics #Traffic Prediction and Management Techniques

paper · pdf · doi:10.48550/arxiv.2012.13478

openalex publication_date 2020/12/24 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28

Abstract

In autonomous driving (AD), accurately predicting changes in the environment\ncan effectively improve safety and comfort. Due to complex interactions among\ntraffic participants, however, it is very hard to achieve accurate prediction\nfor a long horizon. To address this challenge, we propose prediction by\nanticipation, which views interaction in terms of a latent probabilistic\ngenerative process wherein some vehicles move partly in response to the\nanticipated motion of other vehicles. Under this view, consecutive data frames\ncan be factorized into sequential samples from an action-conditional\ndistribution that effectively generalizes to a wider range of actions and\ndriving situations. Our proposed prediction model, variational Bayesian in\nnature, is trained to maximize the evidence lower bound (ELBO) of the\nlog-likelihood of this conditional distribution. Evaluations of our approach\nwith prominent AD datasets NGSIM I-80 and Argoverse show significant\nimprovement over current state-of-the-art in both accuracy and generalization.\n

Related