An official website of the United States government
dot gov icon

Official websites use .gov

A .gov website belongs to an official government organization in the United States.

Secure .gov websites use HTTPS

A lock ( ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites.

Intelligent Transportation Systems
Joint Program Office

Informing infrastructure interventions via novel near-miss data collection

By: VELO.AI, INC.

Project Objective & Overview

Developed and deployed a scalable, AI-driven sensor system to detect, record, and analyze near-miss interactions between vehicles and Vulnerable Road Users (VRUs). The project addresses the critical gap in traditional safety analysis by moving beyond sparse crash data to actionable, proactive near-miss observations.

Problem Statement: Reliance on historical crash data ignores high-frequency "near-miss" events, delaying safety interventions until accidents occur.

Pilot Site: Pittsburgh, PA (Urban and mixed-traffic environments).

Partners: POGOH (Bikeshare Fleet), Carnegie Mellon University Mobility Data Analytics Center (Research), City of Pittsburgh (Advisory).

City-scale predictive safety
City-scale predictive safety
Latest generation edge-AI hardware device for road safety analysis
Latest generation edge-AI hardware device for road safety analysis

Key Results & Findings

Fleet Deployment: Collected 1,000+ miles of data via consumer/bikeshare fleets, capturing confirmed near-misses and collisions.

Predictive Modeling: Built region-wide predictive safety model with CMU, extrapolating localized risk to the full city network.

Advanced Analytics: Validated surrogate safety measures (passing distance/speed) and prototyped 3D event reconstruction.

Infrastructure Assessment: Mapped road surface quality (potholes) and traffic congestion using IMU and vision data.

Challenges: Retrofitting bikeshare fleets proved complex; future scaling requires factory integration.

Video frame overlay of near-miss event
Video frame overlay of near-miss event
3D reconstruction of near-miss event
3D reconstruction of near-miss event

Company Info

VELO.AI, INC. is a Pittsburgh-based robotics company utilizing edge-AI perception and predictive analytics to improve roadway safety for Vulnerable Road Users.

Project POC: Clark Haynes, Founder & CEO
Email: gch@velo.ai
Website: https://www.velo.ai

Next Steps

Commercialization: Actively commercializing technology with fleet and municipal partners to deploy sensor networks for real-time safety auditing.

Advanced Data Products: Developing even more refined datasets, specifically moving toward full 3D modeling of the roadway environment.

City-Scale Trajectories: Analyzing precise actor trajectories (vehicles, pedestrians, cyclists) across entire city networks, including accurate metrics of dangerous road interactions between traffic actors.