Dream Labs opens Dream-1 robot API at $3/hour and exits stealth with $90M seed

So generalized actuator planning and control turns out to be possible. That alone is worth paying attention to.

I just want my meals prepped, the flat cleaned, and the laundry folded. Robots taking over menial tasks would matter for a huge chunk of people.

Menial and gig work isn't going away. If home robots land, those wages get suppressed and the workers can't afford the retraining that was supposed to save them.

Their whole flywheel depends on people using these enabled robots for real work, and from the demos they look unusable for real work. The only toroidal bar of soap I've ever seen was in one of those clips.

This gets acquired by OpenAI for the robotics team, I'm calling it. Starting the countdown timer now.

Why this one and not any of the other dozen robotics startups? Nothing here looks different enough to be an obvious acqui-hire.

I tried one prompt in the playground, move the elephant next to the ball. It moved the ball next to the elephant, then kept pushing both around. Technically the goal state was reached but not the way a human would read the instruction.

That was a fair, simple, very clear prompt. Spending $90M to tweak a model that can't do it in an unvalidated market, I honestly can't tell if I'm out of touch or they are.

The whole post reads as obviously written by an LLM. There's real content in there, but the order and phrasing are hard to follow and it's less convincing because of it.

It's strange to spend years on a product and then publish something an LLM regurgitated. Even a blog post scrawled on a napkin would show that somebody at the company cared.

This lab isn't Anthropic or OpenAI and Dream-1 wasn't a long-awaited release, yet it hit the front page before a single comment. I can find almost no history of anyone mentioning this team.

I posted it after finding the model on robocurve.org, someone on X was saying Kimi K3.1 Pro does badly on that benchmark. It's the first public API I've seen for a random bimanual robot. HN strips "Introducing" from titles, which is why it's just the model name.

It only takes a few upvotes to reach the front page briefly, and on a Saturday there's less competition for slots.

Physical Intelligence's recent API announcement is more impressive. More diversity of robot embodiments and real tasks in actual workplaces.

I can't bridge this to the automation, manufacturing, and warehouse work I've spent my career on. The page is optimized for investors, not customers, and they cite a literal toy as the leading example.

I don't like where this goes. A third party ends up able to control a robot inside my house, and scaling that has huge negative externalities. Hopefully a Chinese lab ships an open source version soon.

How are you going to afford a house when you're competing with $3/hour labor? There are already companies that clean your house for free, they just make the staff wear cameras, and nobody seems to care.

Once there's traction and money in this category there will be open source models too. Xiaomi has one already.

Once a robot does everything an IQ 80 human does, only better and cheaper, that worker is gone. Then IQ 120, then 180. Capital stops optimizing for humans long before the last person is replaced.

This is too ambitious. Get one robot reliably doing things first, and generalize after that.