My career started in risk management, which is like a gift and a curse walking around; you see what’s happening, and you see opportunities. As a consultant for a lot of transportation projects, I saw a gap. With connected and autonomous vehicles being tested, I realized the robots are just one piece of the puzzle but can't do everything and wondered how infrastructure could help.
In 2024 in the US, approximately 39,345 people died from motor vehicle crashes, with more than 6,700 of them being pedestrians, according to the National Highway Traffic Safety Administration. While autonomous vehicles are meant to eliminate human-caused incidents, there are still plenty of questions around how we can make our roads safer and much more quickly than we have in the past. Cities still wait years to assess historical data before making an intersection safer. I envisioned a more proactive solution using infrastructure-to-vehicle communication: sending alerts from our roadways to our cars to warn them about safety conditions in real time.
When I first started in 2018, I didn't know where to begin. Luckily, I found a couple of PhD students experienced in building deep learning models and brought them on to help build the technology. The result is Sensagrate, a computer vision AI platform helping optimize traffic and roadway safety. We provide real-time data on potential incidents, giving cities better eyes on what’s causing them.
Our goal is to communicate with autonomous and connected vehicles based on their location, direction, and speed. For instance, we can alert a specific driver that someone is jaywalking as they approach an intersection. We’re trying to get to a place where, at scale, all vehicles have this form of communication.
Senagrate’s solution includes a suite of products. We conduct near-miss analysis to understand when pedestrians, vehicles, and cyclists potentially collide, helping cities and universities redesign roadways so they’re safer for everyone. We also use human behavior analysis to understand the typical paths people take, which informs safer design concepts.
AI is the core of all this. Getting high-detection accuracy, and at range, is huge for us—like seeing pedestrians crossing beyond the crosswalk at an intersection from 150 meters away at night—so we use LiDAR (Light Detection and Ranging) cameras and radar sensors with our deep learning AI models using TensorFlow and computer vision software. We analyze real-time traffic data and safety events, and then Google Cloud hosts all of our data.