Demonstration of Pishgu: Universal Path Prediction Network Architecture for Real-time Cyber-physical Edge Systems

Abstract

Pishgu is a universal lightweight network architecture for path prediction in Cyber-Physical Systems (CPS) applications, adaptable to multiple subjects, perspectives, and scenes. Our proposed architecture captures inter-dependencies within the subjects in each frame using Graph Isomorphism Networks and attention. In our demonstration, we will show how Pishgu can predict trajectories for pedestrians and vehicles in different real-world scenarios. We will use video data from various sources, such as surveillance cameras, drones, etc., to illustrate the diversity of subjects, perspectives, and scenes that Pishgu can handle. The goal of our demonstration is to showcase the adaptability and robustness of Pishgu in various CPS applications. We will highlight how Pishgu can capture inter-dependencies within the subjects in each frame using Graph Isomorphism Networks (GINs) and attention mechanisms. We will also compare Pishgu with existing solutions on different metrics such as ADE and FDE. We hope that our demonstration will inspire attendees to explore new possibilities for path prediction in CPS applications using Pishgu. In addition, The demonstration will also consist of a poster to highlight the main features of Pishgu and the improvements achieved over the recent trajectory prediction methods.