InfoSalGAIL: Visual Attention-Empowered Imitation Learning ofPedestrian Behavior in Critical Traffic Scenarios

Igor Vozniak

Abstract: The imitation learning of complex pedestrian behavior based on visual input is a challenge due to the underly-ing large state space and variations. In this paper, we present a novel visual attention-based imitation learningframework, named InfoSalGAIL, for end-to-end imitation learning of (safe, unsafe) pedestrian navigationpolicies through visual expert demonstrations empowered by eye fixation sequence and augmented rewardfunction. This work shows the relation in latent space between the policy estimated trajectories and visual-attention map. Moreover, the conducted experiments revealed that InfoSalGAIL can significantly outperformthe state-of-the-art baseline InfoGAIL. In fact, its visual attention-empowered imitation learning tends to muchbetter generalize the overall policy of pedestrian behavior leveraging apprenticeship learning to generate morehuman-like pedestrian trajectories in virtual traffic scenes with the open source driving simulator OpenDS.InfoSalGAIL can be utilized in the process of generating and validating critical scenarios for adaptive drivingassistance systems.