AA-SGAN: Adversarially Augmented Social GAN with Synthetic Data

TL;DR

This paper introduces AA-SGAN, an adversarial data augmentation method that improves pedestrian trajectory prediction by effectively utilizing synthetic data during training.

Contribution

The paper presents a novel adversarial augmentation architecture that enhances the training of trajectory prediction models using synthetic data.

Findings

Significant performance improvements on real-world trajectory prediction.

Effective augmentation of synthetic data enhances model accuracy.

Adversarial training boosts the realism of synthetic trajectories.

Abstract

Accurately predicting pedestrian trajectories is crucial in applications such as autonomous driving or service robotics, to name a few. Deep generative models achieve top performance in this task, assuming enough labelled trajectories are available for training. To this end, large amounts of synthetically generated, labelled trajectories exist (e.g., generated by video games). However, such trajectories are not meant to represent pedestrian motion realistically and are ineffective at training a predictive model. We propose a method and an architecture to augment synthetic trajectories at training time and with an adversarial approach. We show that trajectory augmentation at training time unleashes significant gains when a state-of-the-art generative model is evaluated over real-world trajectories.