Nomadic has raised $8.4 million to fight the wave of autonomous vehicles

To build the autonomous machines of the future, sometimes your brand needs a model.

Companies that make self-driving cars, environmental control robots, or autonomous construction equipment collect thousands, if not millions, of hours of video to watch and analyze.

Editing and uploading these videos is now the responsibility of the people, who have to watch them all. Despite the fast delivery, it does not grow. NomadicMLa startup founded by CEO Mustafa Bal and CTO Varun Krishnan, aims to solve problems for customers who have 95% of their fleet data sitting on the books.

This problem becomes even more difficult when you look at edge devices – the most important ones show rare events and can confuse inexperienced AIs.

Nomadic is working to solve that problem with a platform that turns visuals into sustainable, searchable content through visual language groups. This, in turn, allows for better fleet analysis and the creation of unique datasets to promote learning and rapid iteration.

The company announced an $8.4 million seed round on Tuesday at a total cost of $50 million. The round was led by TQ Ventures, with participation from Pear VC and Jeff Dean, and will allow the company to scale more customers and continue to improve its platform. Nomadic as well got the first prize at the Nvidia GTC competition last month.

The two founders, who met with Harvard computer scientists, “kept solving the same problems over and over again in our operations” at companies like Lyft and Snowflake, Bal told TechCrunch.

“We’re giving people information on their screens, everything that drives AVs (and) robots,” he said. “This is what drives the developers of autonomous systems, not random information.”

Imagine, for example, trying to control an AV’s understanding that it can turn on a red light if a police officer directs it to do so, or isolate itself whenever traffic moves under a certain bridge. The Nomadic platform allows this to be identified for tracking purposes, and fed directly into training pipelines.

Customers such as Zoox, Mitsubishi Electric, Natix Network, and Zendar are using the platform to build intelligent systems. Antonio Puglielli, Senior Vice President of Engineering at Zendar, said that the Nomadic tool allowed the company to expand its operations faster than other outsourcing methods, and that its field expertise set it apart from its competitors.

This type of model-based, predictive tool is emerging as a key driver of physical AI. Data processing companies such as Scale, Kognic, and Encord are developing AI tools for this task, while Nvidia has released a family of open source models, Alpamayowhich can be changed to solve the problem.

Varun argues that his company’s tool is more than a pen; is a “business thinking method: you define what is needed and figure out how to get it,” using a number of examples to understand the situation and put it into context. Nomadic’s sponsors hope to focus on the basics of these standards for success.

“That’s why Salesforce doesn’t build its own cloud and Netflix doesn’t build (a distribution center),” Schuster Tanger, a TQ Ventures partner who led the round, told TechCrunch. “Secondly, an independent company tries to build the Nomadic inside, they get confused by what makes them successful, the robot itself.”

Tanger praises Nomadic’s talent, saying that Krishnan is a world-class chess player who was ranked the 1,549th best player in the world. Krishnan, meanwhile, boasts that all the company’s engineers have published scientific papers.

Now, they are working hard to develop specific tools, such as those that understand the changes in the path from the camera images, or some that find the precise position of the robot in the video. The next challenge, from the point of view of Nomadic and its customers, is to create similar tools for invisible data such as lidar sensor readings, or integrating sensor data across multiple models.

“Analyzing terabytes of video, crunching hundreds of 100 billion-plus parameters, and then finding their correct information, is very difficult,” Bal said.