Genesis AI, a startup that raised a $105 million seed spherical to construct foundational AI for robotics, has unveiled its first style, GENE-26.5, and it comes with marvel palms. In a demo video, the corporate showcased quite a lot of complicated duties carried out through a suite of robot palms it has designed in-house.
“The style has all the time been the objective, as a result of a greater style way higher intelligence,” Genesis cofounder and CEO Zhou Xian instructed TechCrunch. However the corporate quickly learned that it wanted keep an eye on over the {hardware}. “So we made up our minds to move complete stack,” he mentioned.
Different well-funded corporations perform on the intersection of AI and robotics — comparable to Bodily Intelligence and Skild AI. Zhian additionally stated that “there’s most likely 50 or 100 robot hand corporations in the market.” However he and his cofounder Théophile Gervet hope that construction their very own will give them the higher hand.
The important thing distinction is that Genesis’ hand has the similar dimension and form as a human hand — fairly than the two-finger grippers many robotics corporations were the usage of — lowering the space with real-world stipulations.
“That shall we us acquire much more information than used to be prior to now imaginable, to coach a style that may do many extra duties,” mentioned Gervet, a former analysis scientist at Mistral AI who’s now Genesis’ president.
Of all of the bodily manipulation duties showcased within the video under, Gervet’s non-public favourite is cooking, as it proves that the robotic has been ready to finish an extended sequence of adverse duties, comparable to cracking an egg and reducing a tomato. However Genesis has additionally tasked its robots with getting ready smoothies, enjoying the piano, and fixing Rubik’s dice — a robotics gimmick.
Different duties, comparable to lab paintings, are nearer to what might be the economic programs of Genesis’ generation. However what occurs at the back of the scenes is simply as essential: the startup has additionally advanced a sensor-loaded glove that works as a real-life double of its robot hand, accumulating information that may extra readily be used.
“Our thought used to be that if shall we design a robot hand that tries to imitate a human hand up to imaginable, we will in an instant unencumber massive quantities of human information with no need to fret about what other people name the ‘embodiment hole’ in robotics analysis,” Xian mentioned.
Others have attempted their hand at that downside; the principle novelty is how Genesis combines this with its style. The present model is called GENE-26.5 for Might 2026, however Xian expects there might be many iterations, due to the simulation it has advanced. “The actual bottleneck for the iteration pace of the style is analysis. So this is helping us accelerate style coaching so much,” he mentioned.
Past simulation, although, information might be key to coaching fashions that may lend a hand robots carry out extra duties. That’s additionally the place Genesis’ glove may turn out to be useful. Gervet mentioned that, not like clunky information assortment units that get in the best way, it’s only as gentle and simple to put on as the safety gloves already utilized in many industries, whilst reasonably affordable to make.
“We’re in talks with numerous shoppers presently, and numerous the price of a glove can be that, for the primary time, you’ll be able to put on the information assortment software while you’re doing all your day by day task, whether or not it’s a lab technician for pharma or for production,” Gervet mentioned. This is able to even be complemented through ‘selfish video information’ — other people filming themselves doing the duty.
Nonetheless, it is still observed whether or not staff would feel free to put on the very gloves and cameras that might teach robots to exchange them, and whether or not they’re going to get further pay for that coaching. That might be between Genesis’ shoppers and their staff, Gervet steered. “We haven’t nailed the main points but,” he mentioned.
Both approach, they’ll come to a decision to not percentage that information with the startup, the founders stated. However the startup additionally has avenues of its personal to construct its ‘human talent library’ — it would additionally pay third-party companions to gather information. Its style is already educated on “huge quantities of human-based web movies,” in keeping with a press liberate that didn’t point out reimbursement.
Blended with its simulation machine, this would lend a hand Genesis decrease the prices of its generation for real-world programs like the only it has demonstrated. “This marks a very powerful milestone for his or her staff and the robotics trade extra widely,” mentioned Google’s former CEO, Eric Schmidt, who invested within the startup.
In July 2025, only a few months after its advent, the startup had emerged from stealth with a $105 million seed spherical co-led through Eclipse and Khosla Ventures, with further backers together with Bpifrance, HSG, and folks like Schmidt, but in addition Xavier Niel, Daniela Rus, and Vladlen Koltun.
This investment helped Genesis building up its headcount. With workplaces in Paris and California, it has additionally expanded to London. “One large explanation why we made up our minds to be in Europe is there’s a massive ability density throughout the entire continent,” Gervet mentioned. Its staff of 60 other people is divided round “40-45% in Europe and 50-55% within the U.S,” and the startup is these days hiring in all 3 places.
Apart from hiring, the corporate additionally plans to show its first general-purpose robotic in a while, which Xian instructed TechCrunch might be a full-body robotic, now not simply palms. However he insisted that the roadmap remains to be the similar.
“Our objective is to construct probably the most succesful robot machine,” he mentioned.
While you acquire via hyperlinks in our articles, we would possibly earn a small fee. This doesn’t impact our editorial independence.
