What if you could rent a robot’s brain the way you rent a scooter? That’s more or less the pitch from Dream Labs, a San Francisco startup that just opened its Dream-1 robot brain to any developer with an API key. The most powerful version costs $3 an hour, according to SiliconANGLE.
It also comes with $90 million in fresh funding. So is this the moment cheap robots finally get smart? Let’s dig in.
So what actually happened?
On October 19, Dream Labs announced a $90 million seed round led by B Capital and Kindred Ventures. A.Capital, Unusual Ventures, Wing VC, SV Angel, Scribble VC and angel investor Adrian Aoun also joined. That’s a huge number for a seed round, which is usually a company’s first serious money.
At the same time, the company opened its Dream API to the public. An API, or application programming interface, is basically a doorway that lets one piece of software talk to another. Here, you send in a robot’s camera feed plus a plain-English instruction. Then the API sends back motor commands, the low-level signals that make a robot’s actuators move.
The team behind it isn’t new to this. Daisy Cai, a general partner at B Capital, said the team “did some of the defining work in robot learning at NVIDIA.” Joel Jang is the co-founder and CEO. And the startup plans to spend most of the money on compute as it adds more tasks, more robots and more users.
How the Dream-1 robot brain learns
Training a robot usually means collecting data on one specific machine, and that work starts over with each new model. Dream-1 takes a different route. It learns from video of people doing everyday tasks, like stacking shelves or packing boxes. Then it predicts what should happen next and works out the robot action needed to make that happen.
Why video? Because it’s much easier to collect than machine-specific data, so the knowledge can travel across different hardware. In other words, the model learns the task from people first and then maps it onto whatever robot body it’s driving.
Right now, Dream-1 works with two-armed robots that use parallel grippers. Those are the simple two-jaw pinchers you’ll see on a lot of robot arms (we explained how robot grippers work in an earlier post). The company claims it needs no extra data to control a robot like that, even one it’s never seen before. It also tunes the model for each of its 20+ hardware partners.
So far, the tasks stay on the tabletop: picking things up, placing them, and folding laundry. The company says wheeled mobile robots and fully dexterous, humanlike hands come next.
Why this one’s a big deal
Here’s the problem Dream Labs wants to solve. By its own estimate, more than 5 million industrial robots work in factories and warehouses around the world, and most of them do a single task. Teaching one robot one new task can take specialists months. And when you switch to a new hardware model, that work starts over.
Meanwhile, the hardware itself got cheap. “A capable robot arm has gone from the price of a luxury car to the price of a high-end laptop,” Cai told The Robot Report. The robots Dream-1 supports today cost roughly $2,000 to $10,000. For comparison, here’s our breakdown of what buyers really pay for humanoid robots.
So the missing piece is the brain. Jang put it this way: “The hardware is ready. What’s been missing is a brain any robot can use.” He also offered a neat analogy: “When you get into a rental car, you don’t relearn how to drive.” Dream-1 aims to approach a new robot the same way, so it can do useful work “in minutes, not months.”
One early user backs this up, at least anecdotally. Ryan Chan, CEO of MakerMods, which sells $2,000 two-armed robot arms, said his robot “was up and running in a few minutes and attempting tasks we never programmed.”
Wait, $3 an hour?
Yep. According to SiliconANGLE, the most powerful version of Dream-1 costs $3 per hour, while a lightweight version costs 75 cents per hour. The models run in data centers around the world, so the heavy computing happens in the cloud rather than on the robot.
There’s more. If your hardware isn’t supported or you want a new skill, Dream Labs offers fine-tuning as a service. You send a small batch of demonstration data, and a tailored model comes back within hours. SiliconANGLE didn’t list a price for that part.
Our favorite bit, though, is the Robot Playground. Anyone with a browser can type commands to a real robot in the company’s San Francisco office and watch it respond live. No robot purchase needed. Jang said that “instead of edited demos, we’re giving people a working model they can test themselves, mistakes included.” Honestly, that’s refreshing in an industry that loves a slick, heavily edited video.
Okay, but should you get excited?
A little. But let’s keep our feet on the ground. The capability claims are the company’s own, and SiliconANGLE’s coverage didn’t include success rates or benchmark results. So we don’t yet know how often Dream-1 gets a task right on a robot it’s never seen.
The task list is also pretty narrow for now. Picking, placing and folding laundry are useful, sure. Yet they’re a long way from the messy, unpredictable work that real warehouses and homes throw at a robot. Mobile robots and humanlike hands are still on the roadmap, too.
Still, the business model is the interesting part. Dream Labs isn’t selling robots here. Instead, it rents out the intelligence by the hour to anyone who wants it. That feels a lot like how cloud computing changed software, and it could lower the bar for small teams in a big way.
What the Dream-1 robot brain means for smaller players
This is where it gets exciting for places like Armenia. If a $2,000 robot arm plus a $3-an-hour brain can do useful work, then a small workshop, a university lab or a scrappy startup can start experimenting without a giant budget. The Dream-1 robot brain won’t replace a factory automation team tomorrow. But it might let a lot more people try. And with cheap, rentable tools, the folks who start tinkering early usually learn the most.
So here’s the open question we keep coming back to. When robot smarts cost about as much per hour as a coffee, what will small teams build first? We’d love to see the answer come from somewhere other than Silicon Valley.