Red Rabbit Robotics made an appearance at the Humanoids Summit this week – a conference for builders making machines resembling people and selling autonomous labor as a service.
Imagine Amazon Mechanical Turk, but the output is on-premises and physical rather than digital data. In short, Red Rabbit is selling robot workers – initially remote-controlled, but eventually autonomous.
That’s unremarkable, given the long history of factory automation and industrial robots. But just as mainframes gave way to personal computers, Red Rabbit’s ambition comes at a time when robots – empowered by machine learning models – are poised to become personal and broadly capable. They’re ready for roles more challenging than cleaning floors.
The startup, co-founded by Lingkang Zhang (CEO) and David Goldberg (COO), has developed and open sourced a robot called RX1 that consists of a mechanical torso. And if you’re, say, running an assembly line that requires the manipulation of product parts, automating some part of that process may be less cost-prohibitive than it has been.
“We’re building autonomous labor as a service,” explained Goldberg in an interview with The Register. “A lot of companies, either in manufacturing, supply chain, or commercial applications, are constrained by a labor shortage.
“You have a lot of jobs that are repetitive. They’re either dull, dangerous, or dirty, and they have a very difficult time hiring people for that and retaining them – and with those people doing a reliable job. And those people are expensive.
“So the more dull the job is, the more dangerous it is, the more you have to incentivize people to want to do the job, and so therefore, wages go up. Inflation is also probably not helping.
“And so really what we’re trying to do is help fill the labor shortage for those types of jobs, drive the cost of labor down – to half what it is now – while achieving productivity output gains.”
If a robot’s working 24/7, and is three times as productive as a human equivalent working eight hours a day, then you should be able to, for half the cost, get three times the productivity and not worry about the labor shortage
According to Deloitte, 3.8 million new manufacturing employees will be needed by 2033, but “around half of these open jobs (1.9 million) could remain unfilled if manufacturers are not able to address the skills gap and the applicant gap.”
“If you think about it, there’s no reason why these robots can’t work 24/7,” observed Goldberg. “You probably have to figure out power supply and stuff like that, but those things are easy, relatively easy to solve.
“And so if a robot’s working 24/7, and is three times as productive as a human equivalent working eight hours a day, then you should be able to, for half the cost, get three times the productivity and not worry about the labor shortage. So that’s really where we’re trying to go.”
The Register asked Goldberg if he sees this taking away specific jobs, or is it more about removing the most onerous responsibilities faced by employees?
“We see this being an enabler of people being able to focus more on work they want to do and less on work that they have to do – such as moving boxes from place to place, or the same repetitive task over and over again where you develop carpal tunnel syndrome.”
Initially though, a person will still be doing this repetitive motion – just not directly.
Goldberg explained, “We have an existing pilot customer, where our core team is teleoperating the robot at first. This ensures we can refine the UX prior to releasing the software for others to use.”
The plan, he added, is to transition from manual teleoperation to semi-autonomous, to fully autonomous over time.
We asked what it requires in terms of a company maintaining robots. Does Red Rabbit do the maintenance, or does the customer have to hire someone to be their robot wrangler?
“In the long run,” Goldberg told us, “we see this working where there’s two things we can do on that front. So first, we think the customers shouldn’t have to do much from the maintenance standpoint. At the very beginning, it requires us to probably lean in a little bit more – do things that don’t scale that type of approach.”
“And then over time, how can we design the product in a way where it’s so straightforward to basically service the robot that you don’t necessarily need us to come onsite to service it for you? That being said, in the foreseeable future, we will have field support – especially in this earlier stage that we’re in.
“We have some pilot customers now, but I think in the short term, it’s what can we do that just makes it so they don’t have to think about it – even if those things don’t scale for us. Then we’ll figure out how to scale over time.
“If you look at the landscape of the autonomous labor as a service market – which is what we’re going into – a lot of the companies are building for perfecting the tech. You know, a lab setting, things like that.
“And what we’re trying to say is, ‘hey, I think we need to get something working for the customer as soon as possible, and just not perfect everything, but just get the bare minimum going’.
“I think that’s one of the misalignments we see in the industry: everybody here at the Humanoid Summit is talking about why their tech’s better than your tech, right? But at the end of the day, like, who’s out there selling to the customer, solving their problems? And that’s what we’re trying to do.”
Finally, The Reg wanted to know about the data gathering part of the engagement process – how does Red Rabbit generate a feedback loop to improve the product?
Goldberg sees that part of the equation as an organic element of the process. “The way that we see data gathering getting solved over time is, we have to get in-the-wild adoption. It’s all about diverse data, unique scenes, unique scenarios. If you look at the self-driving car market, the more [data] they get on all the possible scenarios that happen on the road, the better the self-driving tech gets. It’s the same thing here with robotics.”
He added that his focus is on utility and adoption – and one of the things that limits adoption is cost.
In reference to the other robots in the exhibit hall, he observed: “If you look around, there’s a bunch of robots here that are a couple hundred thousand dollars each robot, right? Well, we have a pilot customer who wants to order 100, but they’re not going to spend $20 million on the implementation. So we need to drive the cost down to … call it $20,000.
“So that’s what we’re targeting. Then the question is ‘How do you build this in a way that you get a low price point like that, but it’s still reliable?’ And then you get adoption in the wild, and that feeds your data flywheel, and now you can develop further capabilities.”
Maybe the answer to that question is open source. ®
Discussion about this post