Helm.ai said it has commercial partners spanning perception, full-stack autonomous driving, automated data labeling, and generative simulation. | Source: Helm.ai
Helm.ai today said it has signed $70 million in commercial contracts for its foundation models for physical AI over a 12-month period. These contracts spanned global automotive OEMs, Tier 1 suppliers, and industrial automation companies.
The Redwood City, Calif.-based company considers the automotive industry as its flagship deployments, which involve deep partnerships. But, it also has partnerships with customers in mining and construction.
“Those kinds of projects coming to that level of maturity are super exciting for us, because it demonstrates that there’s a real demand for the technology that we’ve built,” Vladislav Voroninski, founder and CEO of Helm.ai, told The Robot Report. “It’s clearly crossing that threshold into actually being deployed in the real world,”
“Also, as a company, we’re now on a path to break even, which is rare in this space, and it’s a testament to the capital efficiency of our approach,” he said.
Founded in 2016, Helm.ai’s software already spans SAE Level 2 through 4 autonomous vehicle (AV) programs, production-track perception in heavy industry, and expanding robotics development.
Inside Helm.ai’s ‘deep teaching’ methodology
Helm.ai trained its foundation models using its unsupervised “deep teaching” methodology to master the structure of the physical world itself. This separates the problem of understanding an environment from the problem of acting in it.
“The way we attack the problem is different from thinking about it as there’s some sensor data in, and a decision comes out,” Voroninski said. “We actually have structure throughout the stack, so each part of the stick is still learned end to end using DNNs [deep neural networks]. The way we think about solving the problem is, first, learn how to perceive the world, and then we actually act on that perception. So, it’s similar to how humans learn.”
Voroninski compared this to a teenager learning to drive. That teenager doesn’t have to drive for millions of hours to experience every single possible scenario they could encounter on the road.
“They can already perceive everything perfectly and make predictions about what the other vehicles or people might do, without actually relying on driving data,” said Voroninski. “That’s a pretty critical thing, and that gives us data efficiency, which is really important in the autonomous driving space, but it’s actually even more important in robotics. I would say it’s essential in robotics.”
An automaker simply has to figure out how they will ramp up collecting data. “For robotics applications, nobody has a large fleet of robots yet. There not really any data available to even train that way,” Voroninski said. “So, data efficiency becomes very, very important.”
“We factor our stack into perception, and then everything downstream from perception,” he continued. “That plays an important role in data efficiency as well as safety certification.”
The result is a system that learns from a fraction of the data, generalizes to environments it has never encountered, and deploys within the compute constraints of real-world physical systems, according to Helm.ai.
AVs give Helm.ai a foundation for scaling to different environments
Voroninski said an important aspect of Helm.ai’s technology is being environment-agnostic. The company’s roots are in autonomous driving, which already requires generalizing across a range of different environments. The system may have to handle a busy city street or a desert road. This provided a good foundation for generalizing across more environments, he pointed out.
“We’ve shown that our technology can generalize to entirely different application areas,” said Voroninski. “We can take the same perception stack and use it for autonomous driving purposes, an open pit mine, or other kinds of industrial environments.”
So far, Helm.ai has projects bound for production in AVs, mining, and construction. It has also applied its AI technology to delivery drones.
“When we train foundation models, we’re not just training on driving data or a specific target area,” Voroninski said. “Our foundation models are trained in a pretty general way across data sets that go well beyond those specific applications, and that’s similar to how a human would learn how to actually do these things.”
Helm.ai learned other things from the automotive industry. Autonomous driving comes with real-time latency constrains and rigorous safety regulations. Being able to handle that industry makes the company well equipped for others, Voroninski said.
Helm.ai’s software is robot-agnostic
Voroninski discussed using Helm.ai’s technology across autonomous vehicles, drones, humanoid robots, and more. While it may seem difficult to generalize to many robot form factors, Voroninski said these different form factors all come with similar problems.
“The perception problem starts with the sensors that you’re using and how they’re configured,” Voroninski said. For AVs, this typically means a 360º view with a camera stack, lidar, and radar.
“We’ve shown that we can work quite well with those types of sensor modalities and simulate all of those at the same time,” Voroninski said. “When you go to other domains, you’re still talking about sensors that are positioned on some form factor in some configuration. From that perspective, it’s exactly the same.”
“It’s essentially the same problem, just different definitions for what you want to detect, what you want to localize, and the kinds of behaviors you’re going to care about,” he continued.
A person on a construction site might behave differently from a pedestrian on the street. But, the same technology can make predictions about both of these people.
Looking ahead, Voroninski said Helm.ai is interested in working with robotics companies across industries and embodiments.
“Where robotics is now is almost where autonomous driving was 10 years ago,” Voroninski said. “It’s really now entering its breakout moment, so we’re just super excited to take things that we’ve learned from the commercial traction we’ve achieved, and apply that to many different areas.”
Register now and help us celebrate 20 years of RoboBusiness!
Tell Us What You Think!