DriveMatrix supervises GenAI to create ‘real’ data augmentations and improve vehicle safety

For most people buying a car, safety is top of mind. That makes it a chief concern for automakers too – but it’s difficult to properly train and validate the vehicle safety features common in advanced driver assistance systems (ADAS) in all conditions when most test drive data is from clear, sunny days. 

Enter DriveMatrix, Cognata’s AI-powered solution designed to improve perception models in ADAS training by generating realistic driving scenarios in diverse conditions. 

DriveMatrix uses supervised generative AI to transform existing test drive images and videos into variations that can replicate diverse weather and lighting conditions. Because the AI-generated videos are layered on top of real data, Cognata is able to augment the data with the highest quality while eliminating known GenAI phenomena’s such as hallucinations, which are a primary concern about the use of generative AI. 

An automaker training and validating an ADAS or autonomous vehicle (AV) can take existing data collected on a clear, nice day – as 70% of data is – and augment it while maintaining fidelity to the original recording, says Shay Rootman, VP of Business Development and Marketing at Cognata, which is based in Israel. That means those critical safety features can be trained to perform in rain, fog or snow, with lane lines faded or removed, on a gravel road or at night.

Because it’s layered on real data, computers also view the augmented data as “real,” which eliminates the domain gap of synthetic data. In addition, the imbalance created because the AI algorithm would otherwise become biased toward that clear, sunny data. 

Cognata Supervised GenAI technology means that the data is repeatable, controllable and predictable – three critical components in the world of generative AI. By generating multiple training scenarios from a single dataset, DriveMatrix helps automakers cut data collection costs, accelerate ADAS development, and improve vehicle safety in challenging conditions.

DriveMatrix builds on and complements SimCloud, Cognata’s autonomous vehicle simulation platform, which was built on AMD Radeon Pro V710 virtual machines hosted in Azure. Automakers continue to use simulation to replicate edge cases – for example, a pedestrian running into the road. DriveMatrix is able to address the coverage problem, and synthetic data fills in where real data can’t exist. With DriveMatrix, automakers can now train ADAS models to react to hazardous weather conditions, low visibility, and complex road environments – areas where real-world test data is often unavailable. This not only enhances vehicle safety but accelerates regulatory approvals and market readiness. 

Cognata leverages Azure’s scalable architecture to transform real test drive data into diverse driving conditions. Using Azure Batch and Azure Scale Sets, Cognata processes vast amounts of data in parallel, accelerating ADAS training while ensuring accuracy. Given the exponential growth of training data, Cognata needed a cloud provider that could scale dynamically. Azure Blob Storage provides the capacity and performance needed to manage massive datasets while keeping costs under control. 

“The process is very heavy on compute,” Rootman says. “Cognata and Azure are meeting in a way that we can really harvest the power of the GPUs with the cloud and the elastic architecture we’re able to leverage from Azure.”

Those GPUs are also from AMD, which has worked closely with Cognata in a collaboration made possible in part by Microsoft. Rootman says he appreciates that Microsoft values its partners, committing to support them with solutions and infrastructure but to never compete directly. Cognata has also had access to customers it otherwise would not have as a small business, thanks to its relationship with Microsoft. 

“It’s not only about compute and it’s not only about not competing,” he says. “It’s about the fact that you’re going to a customer together and winning together.”

Rootman says he is proud that Cognata continues to keep up with the cutting edge of technology. SimCloud was a major breakthrough, but with DriveMatrix, Cognata is pushing the boundaries of generative AI in ADAS training. The company is now exploring new AI capabilities that will further refine vehicle perception and decision-making in real-world driving scenarios. 

“We are finding a way to harvest that power in a way that you can actually control it,” he says. “We are able to take the top technologies and bring them into usefulness with slight modifications.”

Discover more about Cognata and DriveMatrix on Facebook, LinkedIn, X/Twitter and YouTube.