Learn to Quantify Social Interaction with Constraints for Pedestrian Walking

The paper presents "Learn to Cluster," a novel approach for quantifying social interactions among pedestrians in the context of long-term human path forecasting. This probabilistic latent variable generative model learns directly from sequential trajectory observations without requiring labels, thus allowing for scalable integration into trajectory prediction models. Its ability to effectively categorize social interactions enhances the robustness of pedestrian trajectory predictions, which is crucial for autonomous systems such as self-driving cars and social robots.