Driving behavioral learning leveraging sensing information from Innovation Hub

Download Final Report

CAIT project no.: CAIT-UTC-REG46

Fiscal Year: 2020/2021

Status: Final

Principal investigator(s): Xuan(Sharon) Di (PI), Columbia University

Performing organization(s): Columbia University

Coauthor(s): Peter Jin Co-PI, Rutgers University
Yufei Huang, Rutgers University
Zhaobin Mo, Columbia University

Managing organization: Rutgers CAIT

In cooperation with: Middlesex County, Transportation Department,
Partner project manager: Solomon Caviness, Department Head

Supported by: USDOT-OST-R

UTC, grant, or agreement no.: 69A3551847102


The primary goal of this proposal is to develop machine-learning algorithms for driving behavior mining, using real-time vehicle, pedestrian, and infrastructure data. The proposed algorithms will improve our understanding of how people drive on both highways and urban roads, which will help monitor and maintain roadside infrastructure and support the transportation systems to accommodate not only the existing human-driven vehicle but also the upcoming connected and automated mobility systems.

The intended outcome of the project is an algorithm suite to learn human behavior patterns from LiDAR and camera datasets. To facilitate its adoption by public agencies, the software will be open-sourced with friendly interface design. The proposed smart mobility testbed concept could be deployed at local intersections and arterial corridors in the City of New Brunswick, NJ and utilized by the Robert Wood Johnson hospital’s patient shuttle services and parking services.