CS Events

PhD Defense Advancements in Modeling Crowd Navigation from Cognition to Simulation to Prediction
Export Event Save iCal Send to Google Calendar Send to Yahoo Calendar Send to Outlook Live Send to Microsoft Outlook Save iCal Preserve formatting in description (only supported in some calendar applications) Close Friday, September 20, 2024, 01:00pm - 03:00pm

Speaker: Samuel S. Sohn

Location : CBIM 22

Committee: 

Prof. Mubbasir Kapadia 

Prof. Vladimir Pavlovic

Prof. Mridul Aanjaneya

Prof. Dinesh Manocha (External Member)

Event Type: PhD Defense

Abstract: This dissertation advances the intersection of spatial cognition, crowd simulation, crowd flow prediction, and human trajectory prediction, aiming to develop more realistic and efficient models for autonomous systems. The research emphasizes the importance of grounding models in cognitive principles to accurately model human behavior. This approach not only enhances the realism of simulations, but also provides valuable insights for designing safer and more efficient public spaces. However, behavioral realism comes at a prohibitively expensive computational cost. This body of work introduces and significantly improves a novel framework that circumvents this scalability issue by performing long-term crowd flow prediction (LTCFP), transforming the microscopic simulation problem into a macroscopic image-to-image regression problem. We have made foundational contributions to this novel problem in synthetic data generation, feature extraction, modeling, and evaluation. Furthermore, the same cognitive principles that have been used to improve simulation models have proven effective for significantly enhancing the accuracy of LTCFP models. Complementing this macroscopic problem formulation, there is a preexisting microscopic problem known as human trajectory prediction (HTP), which has not yet been operationalized for the same scale of crowds as LTCFP. This dissertation presents new evaluation metrics and datasets to assess the performance of state-of-the-art HTP models, highlighting their limited robustness despite having more specificity than LTCFP models.

: 

Contact  Prof. Mubbasir Kapadia

Zoom Info: https://rutgers.zoom.us/j/94402747359?pwd=kRl2zWaRD1DRwXvNak2twtvPgumJdB.1