I'm not super active in the humanoid robot space anymore, however I did my PhD about 9 years ago in HRI. That was the time of Boston Dynamics, DARPA robotics challenge, and Aldebaran's Pepper and Nao robots.
You mentioned you are building everything open source. What happened with ROS and related projects? Do you build on top of that, or is that all super outdated that a reboot was needed?
Another question I have is: why are you choosing a two-legged human over a four-legged one?
My experiments with two legged robots were mostly bad. Not only did they fall basically all the time but they also had a big drift. So far, I have not seen any large improvements. But again, I might be very outdated.
I always said to my colleagues. The main point stopping robots from picking up is a stable platform. And with the platform I mean walking.
For two legged - I think it will eventually be quite a bit cheaper, it's mostly a software problem to get them to be stable. RL based control has gotten much, much better. For example, I was talking to Trevor Blackwell a few weeks ago, and he was saying the IMU on the original Anybots robot was over $2k, whereas if you throw a noisy IMU into sim, you can get away with something basically from a cellphone. So yea, personally I'm a big believer in needing to solve the robotics intelligence problem, and once you solve that, humanoids will make the most sense as a form factor.
Human feet have metatarsophalangeal joints connecting the toes to the rest of the foot. But humanoid robots don't have these (at least, the vast majority don't). Why? These joints are very useful.
Also, the bottom of the human foot is soft and has thousands of nerve endings. Can we really expect robots to get anywhere near human mobility performance without this level of compliance and sensory sophistication?
In general, I think the best way to think about the differences between human muscles and robot actuators is that human muscles are simultaneously incredible in terms of strength and power density, and also incredibly fragile. Robot actuators are fairly robust, but comparatively poor. Humans are essentially falling apart at all times, but the repair mechanisms in the body do a good enough job that it doesn't matter (although you probably know someone with a "career-disruptive/ending" sports-related injury that shows these mechanisms have limits). Robotics is a long way away from making actuators that can be fixed online in such a process. Even cable stretching in cable-driven mechanisms remains hard to handle effectively, and that's one of the simplest types of mechanism wear.
For walking, the most important thing is that the robot can be simulated well, so in our case, we tried to model the foot contact with the ground in simulation quite accurately. In fact, we found that force sensors in the foot probably help but they're not necessary in simulation, and we wanted to shave off anything that wasn't necessary. I am not sure how close we will get to human mobility - it's definitely a learning process - but you can get much further in simulation than you'd expect.
But this looks like an expensive toy.
The stuff of nightmares is this being adapted by the DoD. I can almost imagine your website as a scene in the prologue of a terminator like movie.
Nightmare 2 is this becomes a companion of some sort. Detroit Become Human goes into this. You have a theme of the robots basically wanting freedom. Which throws out a moral conundrum, if someone buys an AGI enabled bot just to be mean to it, have they done anything wrong.
I like technology , but this feels like step one to a whole lot of weird stuff.
When I click "get in touch" on your github I just land on the website where I can buy the robot. Building the hardware for an autonomous robot is orders of magnitudes easier than the control. Do you think anyone with the capability do develop an autonomus robot will buy this and then just give you the code because its open source?
I agree that the hardware is easier than the software - I am a software guy, personally, but I felt that it was important to do the hardware first so at least we can have a baseline product which we can offer to people. I would personally like to work on this software problem (or rather, build a company to work on this problem), and this seems like the right way to go about funding working on this problem.
It's my hot take that the next big ML breakthrough needs s.th. that learns from its own actions in an environment, so this goes in the right direction imo.
On the other hand a lot of big companies struggle with self driving cars even though they predicted to build this years ago. Also probably all big AI companies work on AI for autonomous robots. Where do you intend to do s.th. different to get a shot at competing with them (when they have so much more capital)?
We haven't updated it much but it's a good starter point
a) some of these definitely look they could be done without hardware or with light hardware support from a staff member?
b) if you do a) and are open source completely I bet you don’t even need to do bounties. The increased participation could mean some great community generated solutions quick.
But on the long-tail of autonomy options? Implementing the latest papers, trying cutting edge solutions, I bet a thriving open-source community could be very helpful ala PR2, given that the hardware is already open-source.
Nothing stopping you from picking and choosing from the various implementations to build into a streamlined product offering on the front end.
I basically think our goal is to solve all the boring stuff and make it work reliably, so that other can people can try out the cool ML stuff more easily
Personally, I think the first real use cases will mostly be entertainment. Humanoids have a high "coolness" factor. Also, I think there's a long tail of random problems which you don't want to buy a new robot to solve, but which, if you have a robot lying around that isn't perfect but is "good enough", might be possible to solve imperfectly. For example, I just had a newborn baby, and I was thinking it would be nice if I had a static arm that could hold his bottle for me. There's a lot of tail end problems like that in your day to day life. But I think the really interesting capabilities will come once there's very good end-to-end models running on-device.
Not sure your research has been through.
The ones that get the most attention from what I've seen are the ones that look female. And the first comment is always about how easy to clean...
All those lonely men spending thousands on the billion dollar revenue generating onlyfans and webcam sites seem to be the immediate consumer market.
Congratulations guys! The technical stuff is above my paygrade, but you have a cracked team and with open source you will have a great chance to be close or at SOTA level at your price point.
However, it looks to me that your core thesis is yes, when the autonomous robots get good enough, even at a medium family car price range they will sell like candies. Sure. But since you also want to have the cash now, to who exactly are you selling? Yes you promise that you will support the full autonomy option, but this sounds weirdly similar to Tesla selling cars promising the FSD, which we all know how that story went.
I'm not saying you won't deliver, I'm just saying you might need to a bit more careful in your story selling/narrative for this. For example, i would be super interested to get one for like 2k if it's not useful now, but paying 10k for essentially promises and possible upgrades is a bit iffy. Hence i would like to at least see some plug in and play current usecases? Even if they are just for fun.
I think this kind of "promise the future, pay now" model does alienates some people, especially when the tech is not ready today. That's why we're open sourcing everything, to avoid the feeling of overpromising on what is ready today. The core idea is that the people who bought FSD early on were very invested in it's success, and that feedback loop is very important for improving machine learning models at scale. The problem happens when actually delivering on the tech takes a long time, but I think we have a fairly clear technical roadmap to make our robot useful. At least, I think there are a lot more intermediate benchmarks for driving value for a humanoid robot than there are for self-driving cars, so I think people who buy it will have a stronger feeling that it is constantly improving.
> that feedback loop is very important for improving machine learning models at scale
Oh will you have your own feedback loop with let's say user's data? Or you meant as an example?
> * That's why we're open sourcing everything, to avoid the feeling of overpromising on what is ready today*
I agree here, it helps the today, but I dont think it helps the feeling of overpromising on what is ready today, its more like, even if it's open source , it does not increase the chances of it being ready/autonomous in the future. (im just playing devils advocate here)
I also agree with the intermediate benchmarks for sure, this is more to what i was referring to, it would be nice to see some more short term usecases/fun applications that are realistic to hit today or in the nearer future, that would drive a lot of sales value, at least for me, rather than go from now to full autonomy. Good luck!
That's more or less the idea - obviously since it's open source we wouldn't scrape peoples' data without their consent, but I would hope that people would contribute to the project in some form. Like, the core idea of the open source ethos is that building something like this collaboratively is a better / cheaper way to scale data collection / experience than us trying to collect all the data ourselves.
> it does not increase the chances of it being ready/autonomous in the future.
Yea that's true. At the end of the day it's just technical execution, so it's pretty risky. I just prefer that if people sign up for something risky, it's pretty transparent what exactly it is they're signing up for :)
Does/will this include the training data, hyperparams, and weights for the models?
People will be reluctant to buy an "open source" robot if the key ML magic to make it work is closed off, e.g. if you charge a subscription for it.
I basically believe that in a world where humanoid robots are actually useful, we will not have any trouble monetizing. Probably we will verticalize manufacturing at some point in the future. I think the bigger risks for our business model are not from people copying us or something, but from not making progress fast enough.
Your current market seems to be "niche toys for rich tech people" and the future market seems very uncertain. I am impressed that you were able to get funding for this idea. How do you get around the "solution in search of a problem (SISP)" objection from VCs? In fact, your founding story indicates that you just liked the technology meaning you had to work backwards to find the business case.
I'm asking because I think many of us would like to get funding for ventures in areas of technology that we are passionate about, but for which the future market potential remains extremely speculative. How do you do it?
I think at volume the price for a good set of hands should settle somewhere around $300-500. Most of it comes down to meeting suppliers where they're at and negotiating mutually beneficial deals. It's not magic but it does require having a good understanding of the hardware in order to negotiate well.
I'm super interested in learning more about the training process of world and robotics model and the data challenges there.
We're all pretty cross-stack - there are some hardware people and some software people, but the product is quite integrated. Personally, my time has been mostly focused on the RL stack recently, and after there are more robots in the wild, I suspect my time will transition to working on building this data feedback loop.
I try to answer questions pretty actively on our Discord so happy to chat there about whatever you like
Just because you can, doesn't mean you should.
It can take years to get a basic bipedal platform operational, and in general it takes 3 times as long to tune the software/firmware. Unless you see actual proof of platform operation for more than a minute, than take any claims as marketing hype.
Could also try a cheap servo hexapod or turtle-bot kit first, as stable well-studied platforms are easier to code on. =3
We are planning to release a similar-size robot later this year (calling it M-Bot) that will be closer in height and price, but our current focus has been on launching the full-size humanoid.
Mechanically, I think Berkeley Humanoid Lite is pretty similar to the 3D printed one we made last year. Our main focus with the K-Bot redesign was to make it not break so much. 3D printed components break a lot and repair time can be quite long. Also, having the wiring routed internally makes a huge, huge difference. So there are some benefits to doing QA on a manufacturing line in terms of quality and consistency.
Also, what’s the different bt the computes: like what’s the onboard computer running (the 2 options)?
Thanks.
For compute - we're exploring a few boards right now, but the base model will be something from Amlogic and the higher end model will be something from Nvidia
Appreciate it
I'm disabled, and one thing I'm really interested in long-term for humanoid robots is disability support work. Disability support work involves a huge variety of individual tasks, as many as a typical person will do in their life, so it's a good fit for an extremely general platform like a humanoid robot. Motorised wheelchairs and dishwashers exist, but a support worker might need to push a wheelchair, do sensitive dishes, do laundry, accurately open and place medications without destroying them, weigh & dose powders, help someone with going to the toilet, cook meals, drive a car, control pets, manage the level of noise/light/smells in the environment to stop someone from being overwhelmed, sanitise surfaces including themselves, navigate confusing interfaces on a phone or computer, help someone drink from a bottle, remember what sort of activities helped a disabled person in the past to be able to do them in the future, help someone with physical fitness activities like punching or kicking a pad, talk to people for someone, carry someone safely in the event of an emergency, make coffee in the morning, monitor intake of various drugs/nutrients/macronutrients, be able to reach and catch someone before they hit the floor if they pass out, help someone walk if they're unsteady on their feet, etc etc. It makes sense to me that it would be cost effective to have one platform which can do all of that with similar performance to a human, rather than automating many of those tasks individually in ways that might not be accessible to some disabled people.
In terms of TAM, absolutely huge amounts of money are spent on disability care (keeping in mind that elder care is also disability care), by both governments and private citizens, and this number is forecasted to continue growing as more people become disabled by COVID-19 and demographic changes increase the elderly population relative to working age adults. As well, there are constantly scandals about how bad conditions are in some area of disability care, almost always due to underpaid, untrained, or unmonitored staff, so there's a lot of demand for both more reliable quality & lower prices; that demand is only going to grow with time. Various government bodies are very large sources of funding that are very concerned with value for money and would pursue any option that could do the job without costing as much - in my country (Australia), there's the NDIS, National Disability Insurance Scheme. They are always looking for ways to consolidate care for less money.
I strongly suspect that any humanoid robot which was good enough to do disability support work would be in extremely high demand in the general population for obvious reasons, as well as being useful as a platform for labour automation, but those are much more speculative. Disability support work is a lot of money for incredibly varied tasks being spent right now. Something to think about.
I do think that this is a great application of a general purpose robot. I'm not sure what the technical timeline will be, but it would certainly be cool for my parents to have such a robot when they are elderly.
Small note: https://www.kscale.dev/why is a 404
Any robotics company that is not thinking about manufacturing from day 1 is not a serious robotics company.