Predictive AI to Save Lives

A huge 131 car pileup, the largest in Wisconsin history, killed a 30-year-old teacher. What could we have done to prevent this? What role could predictive AI play in avoiding these types of accidents? It is time for action and to apply AI technologies to take us to the next level in automotive safety so we can avoid these types of tragedies.

As anyone who has skidded with their bicycle will tell you, when the wheel is turning it’s stopping force is much higher than when it is locked and sliding (see article on static opposed to kinetic friction in rolling motion) . To keep the wheels turning during panic braking, cars typically employ an anti-lock brake system, ABS, which flutters the brakes when they detect wheel locking (YouTube explanation)

In addition to wheel lock detection, modern cars typically keep track of wet/dry road conditions often by listening for the distinctive sound of water blasting in to the wheel wells. Our vehicles also know the tread depth remaining as well as the tire pressure. Other sensors and radios report the temperature outside and their location with GPS. Online there are multiple weather services that can provide hourly weather status. And we even have feeds from services like Waze that tell us about road hazards like car pile ups. We have a wealth of sensors in our vehicles and online available for us to apply to our predictive AI application.

Next let’s turn our attention to our dashboards. They will need work to accommodate these predictive technologies. To get some insight in to this problem I point you to some of the great visualizations that motor sports game designers provide in Xbox and PlayStation (the Forza Motorsport series is an example). The chevron indicators map out the best path on a Heads Up Display (HUD) and turn red when you are going to fast. The Infiniti QX50 has a video explaining their HUD to give you an idea.

So where is this taking us? AI systems are hungry for data and we have mountains of it being generated by a multitude of sensors. We need to enhance, normalize, and analyze these data feeds (Azure, Google Cloud and AWS all have strong analytics offerings that could do this) and get to work training a model that accurately predicts the threshold where the tire will begin to slip. Based on this prediction, we will implement a HUD that keeps our drivers informed when they are approaching the adhesion threshold of their vehicle. 

Drivers will make better decisions when they are in treacherous situations with a predictive HUD. When we finally get to autonomous driving these exact same inputs and models can inform our AI based autonomous systems.

By repackaging this data we can also make a predictive ABS system. We don’t need to flutter. Fluttering gives away precious ground and is sub-optimal. Instead we could be stopping at a rate very close to the physics limit of our equipment.

This technology is already here. It is going to happen. Let’s come together and take what we know and apply it saving lives. I get tired of watching tragedies on the news when I know we can do better. I would love to hear your feedback.