Training Autonomous Vehicles with Synthetic Data: Why Calibrate Invested in Parallel Domain

/ Kevin Dunlap

Autonomous systems such as vehicles, delivery drones, and robots aren’t “smart” from the get-go; they must be trained to operate in the real world, learning how to navigate busy environments and get from point A to B safely. This learning process requires ingesting, labeling, processing, and understanding vast amounts of data. For example, self-driving cars need to analyze video and data from thousands of hours on the road to learn how to navigate safely, and this requires capturing and tagging all this video and data in a time-consuming, error-prone, and costly process.

What if there was a cheaper, faster, more accurate way to quickly provide specialized data tagged at the pixel level? That’s the mission of Calibrate’s newest portfolio company Parallel Domain, whose synthetic data platform enables customers to better train autonomous vehicles by combining real-world and simulated data. Calibrate recently invested in Parallel Domain’s $11 million Series A round, which was led by Foundry Group and supported by Toyota AI, Costanoa, and Ubiquity.

Think about all the nuances humans must understand to drive; a rainy night versus a sunny day, a city street or a country lane, different amounts of streetlights, shadows or glare, the difference between a taxi and a police car, recognizing obstacles in the road, and so much more. Autonomous vehicles must learn these nuances, too.

Parallel Domain creates “parallel” virtual worlds to train autonomous vehicles to understand millions of potential environments, even before they take their first spin. Parallel Domain’s suit of APIs and tools allows developers to create and combine synthetic data into an infinite combination of scenarios to re-create the complexity of the real world. Developers can even train their systems to respond to hypothetical events they may never encounter in the real world, but are nonetheless possible, such as a snowy freeway full of slow-moving semi trucks with a mattress in the roadway.

The company’s addressable market is huge: a group of 30 prominent companies including Google, Apple, and Toyota spent at least $16B in 2019 developing self-driving cars and investment will only increase. Parallel Domain’s platform is already being used by some of the world’s top AI companies to build autonomous vehicles, delivery drones, and delivery robots, and its customer base is growing quickly.

The Parallel Domain team is talented and experienced. The company is led by founder and CEO Kevin McNamara, who previously worked at Apple’s special projects group focused on autonomous systems simulation, at Microsoft Game Studios, and at Pixar Animation Studios. CTO James Grieve has 25 years experience developing games and software systems for Apple, Electronic Arts, and United Front Games, with special expertise in real-time graphics and engineering management. 

We believe the future is bright for autonomous systems, whether it’s self-driving vehicles or aerial drones. But only if developers can train their vehicles safely, quickly, and cost-effectively. Thanks to Parallel Domain’s groundbreaking technology, the fast deployment of fully-autonomous vehicles is finally possible.

At Calibrate, we invest in companies that are creating the future today and we’re excited to work alongside Kevin and the entire Parallel Domain team as they help build the autonomous vehicles of tomorrow.