Seeing Is Believing: Startup Uses 3D Modeling to Make Autonomous Driving Safer

by Nati Amsterdam

It might come as a surprise that only 4 percent of new car buyers, according to a U.K. survey, place safety as a top priority when considering their purchase.

Make that vehicle autonomous, however, and many people become wary of the safety implications of removing the need for a driver.

And yet, each year over a million people worldwide are killed in auto accidents — most caused by human error.

Addressing the heart of this predicament, VayaVision, an Israeli startup that develops software for autonomous vehicles, aims to make self-driving cars safer.

Learning to Adapt in Changing Environments

A child running into the road, a sudden downpour of rain, bright blinding sunshine — rapid changes in the environment are one of the biggest challenges facing autonomous vehicles.

VayaVision’s system helps a car understand, monitor and adapt to such changes. To do this, the company has developed advanced 3D sensing and cognition algorithms, which it runs on the NVIDIA DRIVE PX AI car computer.

The system fuses raw data from multiple sensors, such as lidar, cameras and radar. Running on DRIVE PX, VayaVision’s algorithms process this data to create a precise 3D model of the environment.

The model is based on what VayaVision calls “upsampling” — where additional information is extracted and inferred from frame to frame measurements. These measurements help boost the resolution of the model, providing better detection and enabling the use of less expensive sensors without compromising on safety.

Smarter Cars for a Better Passenger Experience

Traditionally, software for autonomous vehicles has relied on “object-level fusion,” where the data from each sensor is first processed separately before being combined into one integrated view. VayaVision, however, has found that combining the raw data from the car’s sensors before processing, in combination with its “upsampling” technique, leads to more accuracy — with fewer missed detections and fewer false alarms.

VayaVision’s technology could also improve the passenger experience in a self-driving car. While safety is paramount, no one wants to travel in a vehicle that suddenly stops, jerks or makes hard turns due to false alarms.

VayaVision recently presented its work at our inaugural GPU Technology Conference in Israel and fought off fierce competition from other members of our Inception program to become one of five finalists in our Inception Awards. The Inception program has more than 2,000 members worldwide and helps accelerate startups by providing them with access to technology, expertise and marketing support.