Why Cheap Hardware Changes Robotics | Joel Jang
MTS
October 20, 2026
Dream Labs co-founder and CEO Joel Jang discusses coming out of stealth with a $90M seed round to launch Dream 1, a world action model API that enables zero-shot generalization for robotics.
Joel Jang
Back a few weeks ago, Hugging Face released a robot called a Microduck.
In four days, they sold 10,000 Microducks. Wow. But that's $400, it moves around, you know, rattles, it's very cute. But it also has beakers. If you think about it, it can pick up stuff and do useful stuff. Previously, hardware was very expensive because the software layer, the models were weak, but now they're very capable and they can make use of a lot of affordable hardware.
Sophia
Hello, everyone. Welcome to MTS. Today, we are joined live by Joel Jang, who is the co-founder and CEO of Dream Labs. And Dream Labs came out of stealth today, around five hours ago, I believe, with a $90 million seed. So Joel, welcome to MTS.
Joel Jang
Hey, Supyaya, thanks for having me here.
Sophia
Thanks for joining. I know you said that this is one of your first live podcasts and I was just joking. This will be the first of many. You have a pretty interesting background. I mean, you and a bunch of your co-founders, I know you guys were all former NVIDIA researchers, lots of stuff. Tell me a little bit about the origin story of Dream Labs.
Joel Jang
Yeah, for sure. Dream Labs is a company creating robot foundation models. It was created by me along with three co-founders. We used to work together in NVIDIA for about three years together.
We were very bullish on a direction that a new type of model that we started calling world action models. So previous robot foundation models were based off of VLMs, which contains a lot of knowledge about language, but not really about special intelligence. Now we have a new backbone based off of video pre-trained models, and we started calling them world action models, and we were so bullish that we started to start a company, and we are pushing in this direction right now.
Sophia
And why were you bullish? What were you identifying a big need of?
Joel Jang
The thing about world action models is that it uses all videos as training data. So instead of just relying on robot data, all videos including humans doing useful stuff, they learn a lot of priors from those data.
Sophia
Yeah. So tell me a little bit more about the world model landscape. So what is it like right now? How are you guys changing what's actually possible?
Joel Jang
Yeah, I think the biggest thing about the world model landscape in robotics is generalization. So today, we shipped our first flagship world action model called Dream 1
It's the first public API that anyone around the world can use right now. It generalized zero-shot to unseen robots, unseen tasks, and unseen environments. So anyone can just use the API to control their robot right now.
Sophia
So anyone can use their API to control their robots. I'm curious what this actually just means for the robotics industry. So if you have a robot and you, what does the current landscape look like? I'm assuming a lot of people use generalized models, things that aren't necessarily designed specifically purpose-built for robots. Now that you have an API, so anyone can just use that today inside of their robots. What capabilities does that give them that they didn't have before?
Joel Jang
Yeah. Thank you for asking me this question. So let's take a journey to the journey of a robot developer. So first, they have to buy a robot, and then they have to choose a task that they want to automate, collect data, choose any open-source robot foundation model, including Pi, Groot, rent GPUs, train, and rent GPUs to host a model for inference, right? There's a lot of steps, so a lot of friction. And for a single task, it takes like three months to six months just to get to a high enough success rate to start at that point.
Dream Labs aims to remove these layers. So now, zero-shot robot generalization, task, and environment generalization works. So they can just start using the model.
Now, they still have to buy a robot. So that's why we've been very rigorously focused on providing strong capabilities for everyday robots. So the emotional promise that we make to our users is, we provide everyday capabilities for everyday robots.
Sophia
Everyday capabilities for everyday robots. Where does someone just go and out and buy a robot? Is that a thing that people in the mainstream are doing, and then you just hook it up, you connect it to this API, and now it has all of these capabilities just straight out of the box?
Joel Jang
Read the other 10 minutes of this episode here.
$15 gets you this episode and 99 more. Credits never expire. After checkout, one link brings you back here to read it.
Not what you expected? Full refund within 14 days.
Thousands of transcripts fetched by people building searchable podcast archives
Fetch transcripts with code
The same credits work from a script or an AI agent. The demo key returns a sample episode in full, no card needed:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.
From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.
Using your own key:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/1000797195040