Proactive Robot Movements in a Crowd by Predicting and Considering the Social Influence

Dense crowds are challenging scenes for an autonomous mobile robot. Planning in such an interactive environment requires predicting uncertain human intentions and reactions to future robot actions. Concerning these capabilities, we propose a probabilistic forecasting model which factorizes the human motion uncertainty as follows: 1) A (conditioned) normalizing flow (CNF) estimates the densities of human goals. 2) The density of trajectories toward goals is predicted autoregressively (AR), where the density of individual social actions is inferred simultaneously for a dynamic number of humans. The underlying Gaussian AR framework is extended with our SocialSampling to counteract collisions during sampling. The model allows us to determine a crowd prediction conditional on a particular robot plan and a crowd prediction independent of it for the same goals. We demonstrate that the divergence between the two probabilistic predictions can be efficiently determined and we derive our Social Influence (SI) objective from it. Finally, a model-predictive policy for robot crowd navigation is proposed that minimizes the SI objective. Thus, the robot reflects its future movement in order not to disturb humans in their movement if possible. The experiments on real datasets show that the model achieves state-of-the-art accuracy in predicting pedestrian movements. Furthermore, our evaluations show that robot policy with our SI objective produces safe and proactive behaviors, such as taking evasive action at the right time to avoid conflicts.