SoPhie: An Attentive GAN for Predicting Paths Compliant to Social and\n Physical Constraints

Abstract

This paper addresses the problem of path prediction for multiple interacting\nagents in a scene, which is a crucial step for many autonomous platforms such\nas self-driving cars and social robots. We present \\textit{SoPhie}; an\ninterpretable framework based on Generative Adversarial Network (GAN), which\nleverages two sources of information, the path history of all the agents in a\nscene, and the scene context information, using images of the scene. To predict\na future path for an agent, both physical and social information must be\nleveraged. Previous work has not been successful to jointly model physical and\nsocial interactions. Our approach blends a social attention mechanism with a\nphysical attention that helps the model to learn where to look in a large scene\nand extract the most salient parts of the image relevant to the path. Whereas,\nthe social attention component aggregates information across the different\nagent interactions and extracts the most important trajectory information from\nthe surrounding neighbors. SoPhie also takes advantage of GAN to generates more\nrealistic samples and to capture the uncertain nature of the future paths by\nmodeling its distribution. All these mechanisms enable our approach to predict\nsocially and physically plausible paths for the agents and to achieve\nstate-of-the-art performance on several different trajectory forecasting\nbenchmarks.\n