2024 Robotics Year in Review

Robotics finally feels like it's happening

If you’re reading this, I probably don’t need to tell you that 2024 was a huge year for robotics. As a long-time robotics researcher, it’s been amazing to watch; some of the things that I always dreamed about actually seem to be happening.

For me, there are three big stories:

1. The proliferation of amazing robotics startups: we have Physical Intelligence and Skild raising eye-watering $300M+ rounds, but also really cool companies like Hillbot, Watney, Reflex Robotics, and more.

2. Even university researchers are leaving the laboratory and entering the real world: papers like Building-Wide Mobile Manipulation, Ok-Robot, and WildLMA showed complex, exciting robot behavior happening in challenging, unstructured environments.

3. Hardware is getting cheaper and is broadly available: Stretch RE3, of course, launched this year. Unitree has made amazing robots for a while, but even they now have a ton of real strong competition. Booster Robotics, etc. A new generation of “affordable robot companies” like Innate and Hackerbot.

The Startups

From Hillbot: a mobile manipulator trained to perform pick-and-place tasks from sim to real. Blog post

The startup scene exploded this year. I mentioned Hillbot, Watney, Reflex Robotics, Skild, and Physical Intelligence, but these are hardly the only companies launching, raising, or expanding their scope dramatically. The Bot Company - started by Cruise and Tesla veterans - raised $150 million. Cobot announced their new robot, which is already being used in logistics and healthcare. YC 2024 was filled with cool robotics startups, like K-Scale Labs with their low-cost humanoid, and Weave Robotics building home service robots. And there’s an ecosystem of cool companies building tools for roboticists, like ReSim, Rerun.io, and Foxglove (who put on the very cool Actuate conference this year). I’m certainly missing a lot of things.

Simulation-based learning is one of the core technologies enabling this. Hillbot, for example, is making heavy use of sim-to-real. Skild’s amazing valuation is predicated on sim-to-real for training general-purpose embodied intelligence. Another one to watch is Lucky Robots, who builds maybe the best-looking simulator around right now:

And you can just download this: https://github.com/luckyrobots/luckyrobots

So, right now, if you want to get started with simulation, you have a few really good options: ManiSkill3, Lucky Robots, and of course NVIDIA Omniverse. Just an example of how the ecosystem around robotics startups is getting stronger as well.

Robotics Research in the Real World

This is the year where robotics research really started to get outside the lab, and I’m here for it. In my opinion, this is long overdue: robotics research in controlled, laboratory environments can only teach us so much, and in my opinion we should really stop doing this kind of research.*

I highlighted a few of these above: Building-Wide Mobile Manipulation, Ok-Robot, and WildLMA, and wrote a bit about this in an earlier blog post as well.

Paper Notes: Building-Wide Mobile Manipulation

Also worth noting was the amazing dancing humanoid robot from Xialong Wang’s team at UCSD (paper link). Seeing this kind of research done with lower-cost, general-purpose robotics hardware is a very new development and a thrilling one.

* Note to PHD students: not career advice. Keeping your robot in your lab makes it easier to have your method look good, which makes it easier to get a job.

Hardware Availability

Booster T1 robots participating in Robocup 2024. Small, low-cost humanoids have started to appear everywhere; the proliferation of great options and strength of competition will make robotics research much stronger going forward. Source

At ICRA 2024, Unitree Robotics announced its G1 humanoid for about $16,000 USD. This changed everything.

Yes, the price is misleading. That’s for a robot that’s essentially a 16k remote control car; it has no API access, you can’t build with it. Stretch is still the best mobile manipulator for developers.

But it’s set the tone for robotics hardware development. EngineAI announced a $13,000 humanoid. Booster Robotics is selling humanoids for about $30,000, but that actually seems to include API access — they’ve even become an official partner of Robocup.

There’s also a set of low-cost companies coming out. The Huggingface LeRobot project now supports the cheap Koch arms ($237 as of this post), and now has their own, even cheaper, SO-100 arm as well. Innate is building a ~$2000 robot for hacking with AI and LLMs. Hackerbot, likewise, is building a home robot for about ~$1,300, targeting developers and hackers.

The Hello Robot Stretch RE3 launched earlier this year as well, and is still by far the best mobile robot for in-home use at $25,000. It’s got onboard compute, fully open software (including an AI stack), and is roughly human height - unlike these cheaper, unlaunched products. Still, I’m glad to see that the space is expanding.

Looking Forward

Many robotics companies are trying to deploy real-world systems at scale in 2025. Some will succeed. Source

It’s easy to be optimistic about the future of robotics right now, but there are plenty of risks too. Many, many companies have promised huge breakthroughs in 2025.

The big one, obviously, is AI. We’ve seen partnerships like Figure and OpenAI, and Apptronik + Google Deepmind. I don’t think robotics will scale without AI; traditional robotics requires quite a lot of manual engineering effort which is very difficult, very expensive, and very brittle. By contrast, using foundation models to power AI systems makes developing robotics stuff very simple. You can check out my own Stretch AI for a proof-of-concept system!

Introducing Stretch AI

So, the fundamental pieces we need for general-purpose robotics, in my opinion, are all in place. But for this to actually work, we still need to figure out how to train these things, how to build scalable world representations for general-purpose robots, how to train these world representations and fine-tune them. Will we do all of this in 2025?

Probably not. There’s a lot of work ahead of us.

Still, I think it’s better to be optimistic. The optimists will get the timelines wrong. 1x and Weave may not launch consumer home robots this year. But they’re directionally correct; this is going to happen, and it will surprise people when it does.

Let’s take, as an example, self-driving cars, something I wrote about recently at length. Waymo offered millions of rides in their fully-autonomous taxi service last year, and plans to deploy in Tokyo early this year. There is talk of their unit economics starting to work, meaning they may actually have a real business after spending $10 billion.

Almost 10 years ago — in 2016 — I was an intern at Zoox. Self driving was hyped and felt inevitable. But most of the big, serious players from back then - Waymo, Zoox, and Tesla - are still in the game now (Cruise being the obvious exception). One company I was overly cynical about (Wayve) is still doing great. Everyone’s estimates were wrong as to timing - some too pessimistic, many too optimistic. But we got there, and in 2024 you can get in a robot taxi and have it drive you around.

So, my final thoughts:

• Many people are making predictions about what will or won’t happen in robotics in 2025. Basically all of these people will be wrong.

• Robotics will continue to scale up, to deploy into the real world in new and interesting ways.

• This is a “make or break” year for many companies: large scale robotics deployments need to happen and need to work.

It’ll be exciting to be a part of.

If you like these posts please subscribe and leave a comment!