Research

LiDAR Perception for Autonomous Mobility

Light Detection and Ranging (LiDAR) is a remote sensing method that uses LASER (Light Amplification by Stimulated Emission of Radiation) pulses to accurately measure range (variable distance) information by measuring the time for the reflected light beam to return. It finds its applications in many areas including land and structural surveying, powerline inspections, forestry, farming, mining, and quite intuitively in intelligent transportation systems such as intelligent driver aid systems all the way to fully autonomous driving. Recently, URBANITY Lab started research in the LiDAR perception domain with a goal to improve pedestrian and work-zone safety applications. The research focuses on developing real-time 3D object detection and identification algorithms for point-cloud data received from LiDAR devices to maximize real-world applicability. Please scroll down to our video section below to find more.

LiDAR Perception – Detection and Identification in Point Cloud Data

Applications of Koopman Operator in Intelligent Transportation Systems

Koopman mode decomposition is a strong tool for analyzing nonlinear dynamics. Koopman operator is an infinite-dimensional operator which transforms nonlinear dynamics into linear one in its observable's space. The jeopardy of computing this infinite-dimensional operator has successfully been solved by taking its finite-dimensional approximation using purely data-driven algorithms such as – DMD (Dynamic Mode Decomposition) algorithm. Even though the application of Koopman operator in fluid mechanics community is well known, its vast potential in analyzing traffic dynamics is not fully explored. Currently, we are investigating all those possibilities. The eigen values of Koopman operator of NGSIM data set is shown beside.

Physics Informed Deep Learning

Physics informed deep learning (PIDL) equips a deep learning neural network with the strength of the governing physical law of the learning objective. It can be applied in simulating fluid dynamics, predicting vehicle trajectories, estimating traffic conditions, and so on. One application scenario – traffic state estimation (TSE) – has embedded challenges because of the sparsity of observed traffic data and the sensor noise present in the data. Our case study shows encouraging results on the accuracy and convergence-time of the PIDL algorithm for varying levels of scarcely observed traffic density data — both in Lagrangian and Eulerian frames. It demonstrates the capability of PIDL in making accurate and prompt estimations of traffic states.

Videos

Signal free intersection

We develop a framework of signal free intersection to demonstrate the throughput efficiency involving autonomous vehicles and the control architecture regarding the conflict points of vehicle passing the intersection from different directions and various maneuvers (left and right turns, crossing). Below videos demonstrate the throughput difference of a signal free intersection developed using the proposed framework and established intelligent driver model (IDM) and a signalized intersection under the same traffic condition.

Custom Point Cloud Data Visualization

It is our tool to visualize point cloud data in near real-time to easily analyze the high-level clarity of data. It is necessary to analyze data on a high-level in order to ensure that the data is ready to be processed further or to be fed to deep learning pipelines.

Deep Learning Model Analysis – PIXOR

We used our LiDAR data on an object detection model (PIXOR) trained on KITTI dataset (http://www.cvlibs.net/datasets/kitti/). It is an analysis demo showing that generality of deep learning models suffers due to the shifts in elevation of LiDAR device.

Small Scale CAV Test-Bed

• A small-scale test bed for testing potential ITS applications.

• A hybrid approach bridging gap between theory, simulation and practice.

Posters Presented by Team Urbanity

Past Research

Modeling Information Spreading on Social Media

Understanding the spread of information is critical for the development of governmental regulations and public policies that rely on popular participation and sentiment to succeed. Inspired by well-established modeling frameworks – ignorant-spreader-recovered model for information spread and Sethi model for marketing – we propose a new model designed for illustrating the information spread on social media.