Social-BiGAT: Multimodal Trajectory Forecasting using Bicycle-GAN and\n Graph Attention Networks

Abstract

Predicting the future trajectories of multiple interacting agents in a scene\nhas become an increasingly important problem for many different applications\nranging from control of autonomous vehicles and social robots to security and\nsurveillance. This problem is compounded by the presence of social interactions\nbetween humans and their physical interactions with the scene. While the\nexisting literature has explored some of these cues, they mainly ignored the\nmultimodal nature of each human's future trajectory. In this paper, we present\nSocial-BiGAT, a graph-based generative adversarial network that generates\nrealistic, multimodal trajectory predictions by better modelling the social\ninteractions of pedestrians in a scene. Our method is based on a graph\nattention network (GAT) that learns reliable feature representations that\nencode the social interactions between humans in the scene, and a recurrent\nencoder-decoder architecture that is trained adversarially to predict, based on\nthe features, the humans' paths. We explicitly account for the multimodal\nnature of the prediction problem by forming a reversible transformation between\neach scene and its latent noise vector, as in Bicycle-GAN. We show that our\nframework achieves state-of-the-art performance comparing it to several\nbaselines on existing trajectory forecasting benchmarks.\n