Genesis AI, a French robotics-focused startup, has taken a significant step forward in the field of embodied artificial intelligence. The company recently announced the release of its first foundational model, GENE-26.5, and unveiled a human-like robotic hand built entirely in-house. The hand, which closely resembles a human appendage, is designed to perform complex and delicate tasks ranging from cooking and preparing smoothies to solving a Rubik's Cube.
The announcement marks a major milestone for the startup, which completed a $105 million (approximately £77.2 million) seed funding round in July of last year. According to co-founder and chief executive Zhou Xian, the company initially focused on developing a model to power real-world robotics applications. However, it quickly became clear that to achieve precise control over the entire system, building a dedicated robotic hand was necessary. “We wanted to have full control over the hardware and software integration,” Zhou Xian explained. “By designing our own hand, we can ensure that the data we collect and the actions the model takes are perfectly aligned.”
The field of robotic manipulation has long been constrained by the limitations of grippers and simple claw-like end effectors. Many robotic hands developed by other companies have only two or three fingers, which makes it difficult to replicate human dexterity. Genesis AI’s hand, by contrast, is anthropomorphic in its design, with five fingers and a palm that bear a striking resemblance to a human hand. This design choice offers a significant advantage: data collected from human demonstrations can be more directly transferred to the robotic device. When a human hand performs a task, the resulting movements, grip patterns, and tactile feedback can be mapped more intuitively onto a robot hand that shares the same kinematic structure.
To further bridge the gap between human demonstration and robotic learning, Genesis AI has also developed lightweight sensor gloves. These gloves are intended for use by workers in fields such as pharmaceuticals, manufacturing, and other industries where intricate manual tasks are common. When worn, the gloves capture detailed data on hand movements, finger positioning, pressure, and force applied. This data can then be fed into the AI model, allowing it to learn from real-world human dexterity.
However, the initiative raises a broader question: would workers feel comfortable wearing gloves that might ultimately help train a robotic replacement for their own jobs? Zhou Xian acknowledged this concern, but he noted that companies adopting such technology are often more focused on augmenting human workers rather than replacing them entirely. In many manufacturing settings, robotic systems can handle repetitive or ergonomically challenging tasks, freeing human workers to focus on higher-level responsibilities that require judgment and adaptability. Still, the ethical and labor implications of data collection in the workplace are likely to remain a topic of debate.
Genesis AI said that data from the gloves can be combined with information gathered from videos of workers performing tasks. The integration of motion capture data with visual observations provides the AI with a richer training dataset. This multimodal approach is becoming increasingly common in the development of foundation models for robotics, as it allows the system to generalize across different environments and tasks. By learning from human demonstration videos and sensor data, GENE-26.5 can acquire a broad range of manipulation skills without requiring explicit programming for each individual action.
In a demonstration video released by the company, the Genesis AI hand was shown carrying out an array of impressive feats. In one scene, a pair of hands worked in coordination to prepare a smoothie, carefully selecting ingredients, placing them in a blender, and operating the controls. In another clip, the hands displayed remarkable dexterity by playing a piano, confidently striking individual keys in a melodic sequence. These demonstrations are intended to highlight the hand’s ability to execute fine motor skills that have traditionally been challenging for robots. The Rubik’s Cube solution, for example, requires not only precise finger movements but also the ability to rotate the cube through a series of complex manipulations while maintaining a secure grip.
Founded with a mission to accelerate the development of robots that can operate safely and effectively in human environments, Genesis AI has deep roots in both Europe and the United States. Co-founder and president Theophile Gervet previously worked at Mistral, the prominent French artificial intelligence company. The company retains a substantial presence in Europe, a decision that its leadership attributes to the continent’s rich talent pool and the dense concentration of potential industrial customers. European manufacturers in sectors such as automotive, aerospace, and pharmaceuticals are increasingly exploring robotics and automation to address labor shortages and improve productivity.
Genesis AI’s current global team consists of around 60 people, split roughly evenly between the United States and Europe. The largest contingent is based in San Carlos, California, where the company’s headquarters is located. It also maintains offices in Paris and London. This transatlantic structure allows the startup to tap into the innovation ecosystems of Silicon Valley and Europe’s leading research institutions simultaneously.
The founders’ vision extends far beyond a mere robotic hand. The company has confirmed that it is actively working to integrate its mechanical hand into a full-body, general-purpose robot. Such a humanoid robot would be capable of navigating and interacting with the world in ways that more closely mirror human behavior. Zhou Xian envisions a future where robots can assist with household chores, provide care for the elderly, or take over dangerous jobs in disaster response and industrial maintenance. The journey toward that future, however, is a long one. While foundation models in natural language processing have advanced rapidly, the field of robot learning is still grappling with the complexities of real-world physics, sensorimotor control, and the unpredictability of human environments.
One of the critical challenges in building general-purpose robots is the so-called data bottleneck. Training an AI to understand language or images can be done by scraping billions of records and photographs from the internet. For robotics, however, the data must include physical actions and their consequences. Collecting this data is far more expensive and time-consuming. Genesis AI’s emphasis on human sensor gloves and video learning is an attempt to circumvent this bottleneck by leveraging existing human activities. As factories and laboratories already have workers performing skilled tasks, instrumenting them with lightweight gloves and cameras can produce valuable training data at a lower cost than manually teleoperating robots for each new skill.
The startup’s seed funding round, one of the largest for a robotics company at that stage, underscores investor optimism in the sector. The round was co-led by prominent venture capital firms Eclipse and Khosla Ventures. Additional backers include Bpifrance, the French public investment bank; HSG, an investment firm known for its focus on deep tech; Eric Schmidt, former CEO of Google; Xavier Niel, a telecom entrepreneur and prominent French business figure; Daniela Rus, director of the MIT Computer Science and Artificial Intelligence Laboratory; and Vladlen Koltun, an Apple distinguished scientist. With such high-profile support, Genesis AI has positioned itself among a handful of startups racing to develop humanoid robots that can operate in real-world settings.
Investor interest in humanoid robots has surged in recent years. Companies like Tesla, Figure AI, and Agility Robotics have all announced plans for humanoid machines. While the hardware has advanced considerably, the software—the artificial intelligence that drives the robot—remains the key differentiator. Genesis AI aims to distinguish itself by focusing on dexterous manipulation rather than mere locomotion. A humanoid robot that can walk but cannot use its hands is of limited practical use. By mastering the hand, Genesis believes it can unlock a wide range of applications in industries, logistics, and even domestic care.
The company’s progress in creating a high-fidelity robotic hand also raises expectations among researchers in the field of embodied cognition. Some experts believe that having a body closely resembling human form may be essential for AI systems to learn certain concepts, such as tool use or intuitive physics. The Genesis AI hand, with its human-inspired design, could serve as a platform for experimenting with these ideas in ways that less anthropomorphic grippers cannot.
Still, skepticism remains about the practicality of such technologies in the near term. Robotic hands with high degrees of freedom are notoriously difficult to control. Hardware fragility, sensor noise, and actuator limitations can all hamper performance. Moreover, reproducing the tactile sensitivity of human skin is a major engineering hurdle. Genesis AI has yet to disclose the full specifications of the hand, including how many degrees of freedom it possesses or the type of tactile sensors embedded. Industry observers will be watching closely to see how the hand performs in more rigorous, untested scenarios beyond staged demonstrations.
As for safety, the company has stated that its design prioritizes safe physical interaction, which is crucial if these hands are to work alongside people. The compliance of the joints and the force exerted at the fingertips must be carefully controlled to prevent injury. Future iterations are expected to incorporate more advanced compliance and force-feedback algorithms. The collaboration between the US and European teams will likely accelerate this development, as both regions offer complementary expertise in artificial intelligence and precision engineering.
The path to full-body robots will require significant further innovation. Data collection, model scalability, computational efficiency, and hardware durability are all areas where Genesis AI will need to advance. But with a strong team, a considerable capital injection, and a first product already in hand, the company appears well-positioned to make further strides.
As the field of artificial intelligence continues to evolve, the convergence of large foundation models with physical robotic platforms represents one of the most exciting frontiers. GENE-26.5, the model unveiled by Genesis AI, is among the first of its kind designed specifically for robotic control. Unlike language models or image generators that output sequences of tokens, GENE-26.5 is meant to generate sequences of actions. This distinction is essential: a machine that can understand the world but not act upon it is fundamentally limited. Genesis AI seeks to bridge that gap by giving its model a highly capable body.
In the short term, the company plans to continue refining its foundational model, making it more robust and efficient. It also aims to expand its data acquisition efforts by deploying the sensor gloves in additional industries. The lessons learned from the robotic hand are expected to inform the design of the full-body platform. While the company has not set a public timeline for that project, the momentum is unmistakable. In an industry where incremental progress can take years, Genesis AI is moving at what many observers consider to be a rapid pace.
With offices in three major tech cities, a team split across two continents, and deep institutional support from top investors, the sum of its ambitions is clear. The eventual release of a full humanoid system would be a groundbreaking achievement, but it is by no means a forgone conclusion. The challenges are immense, from the mechanical complexity of moving a bipedal corpse to the AI complexities of navigating cluttered and unpredictable environments. Genesis AI, nonetheless, has started with a crucial component: the hand. How far that hand can reach will determine not only the company’s fate, but perhaps the future of dexterous robotics as a whole.
Source: Silicon UK News