Genesis AI Unveils GENE-26.5 and a Human-Like Robotic Hand
French robotics-focused start-up Genesis AI has released its first foundational model, GENE-26.5, together with a human-like robotic hand developed in-house. The company says the hand can carry out complex and delicate tasks, including cooking, preparing a smoothie, playing the piano, and solving a Rubik's Cube. The launch marks a notable step for a young company that is trying to build both the artificial intelligence software and the physical hardware needed for advanced real-world robotics.
Genesis AI completed a $105 million seed funding round in July of last year. That early capital raise gave the company room to pursue an ambitious full-stack strategy. Instead of relying on off-the-shelf robotic hands, Genesis built its own hand so it could exercise more precise control over the entire system. Co-founder and chief executive Zhou Xian said the company began with a focus on a model to power real-world robotics, then moved into hardware to improve accuracy and integration across the stack.
A Hand Designed to Mimic Human Anatomy
Many robotic hands on the market use two or three fingers. Those designs can be effective for specific industrial grips, but they often limit the range of tasks a robot can learn. The hand from Genesis AI closely resembles a human appendage. That similarity matters because data collected from human demonstrations can be transferred more directly to the machine. A robot with human-like joints and proportions can potentially learn from videos, gloves, and teleoperation sessions with less translation error.
Genesis has also created lightweight sensor gloves that workers can wear in fields such as pharmaceuticals or manufacturing. The gloves capture motion and force data that can be passed along to the AI model. The company said data from the gloves can be combined with information gleaned from videos of workers carrying out tasks. This multimodal approach could help the model generalize across different environments and objects. At the same time, the strategy raises a sensitive question: whether workers will be happy to help train a potential robotic replacement. Companies deploying such systems may need to address concerns about job displacement, data ownership, and workplace surveillance.
Demonstrations Show Delicate Dexterity
In a demonstration video, Genesis showed a pair of its hands carrying out complex or delicate processes. The tasks included preparing a smoothie and playing the piano. Such demonstrations are designed to highlight fine motor control, hand-eye coordination, and the ability to handle objects that are not rigid or predictable. A smoothie involves ingredients, tools, and containers that can shift, spill, or require different amounts of pressure. Playing the piano requires precise timing and independent finger movement. Solving a Rubik's Cube adds another layer: the hand must manipulate a jointed object through many small, sequential turns.
These tasks are not necessarily the same as commercial deployment. A robot that can perform a staged demonstration may still struggle in a cluttered kitchen, a busy hospital, or a factory floor where lighting, noise, and human movement vary. Nevertheless, the video signals the direction of travel for Genesis AI: general-purpose dexterity rather than narrow automation.
Leadership, Offices, and European Roots
Co-founder and president Theophile Gervet previously worked for France's Mistral, a prominent European AI company. Genesis retains a substantial presence in Europe. The company attributed that presence to the continent's large talent base and its potential market of industrial customers. Its team of about 60 people is split roughly evenly between the United States and Europe, with a slightly larger group at its US base in California. In addition to San Carlos, California, Genesis has offices in Paris and London.
The transatlantic structure reflects a common pattern in modern AI and robotics. European founders and researchers often maintain strong ties to labs, universities, and industrial partners across the continent. At the same time, California offers access to venture capital, engineering talent, and customers in technology and advanced manufacturing. By keeping a foot on both sides, Genesis can recruit from a wider pool and work with European industrials that are eager to automate.
Full-Body Robot in Development
Genesis said it is working to supplement the mechanical hand with a full-body, general-purpose robot. That ambition places the company in a growing field of developers pursuing embodied AI. A hand alone can be a powerful data collection and manipulation platform. A full humanoid or mobile robot, however, would need to integrate locomotion, balance, perception, power management, and safety systems. Each of those areas brings additional engineering challenges and cost.
The company's seed funding round last year was co-led by Eclipse and Khosla Ventures. Additional backing came from Bpifrance, HSG, former Google chief Eric Schmidt, telecoms entrepreneur Xavier Niel, MIT Computer Science and Artificial Intelligence Laboratory director Daniela Rus, and Apple distinguished scientist Vladlen Koltun. That investor list combines institutional venture funds, European public investment, and well-known individuals in AI and robotics. It also suggests that Genesis has been able to attract attention despite competition from better-funded humanoid robot companies.
Why Foundation Models Matter for Robotics
Foundation models have transformed language and image processing by learning from massive datasets and then adapting to many downstream tasks. Robotics researchers are trying to replicate that success. The challenge is that robotic data is far harder to collect than text or images. Every movement happens in physical time, with wear and tear, safety constraints, and variations in sensors. A foundation model for robots must learn not only what to do but also how to move safely and reliably in the real world.
GENE-26.5 is described as Genesis AI's first foundational model. Its role is to power real-world robotics, likely by learning representations of manipulation tasks and translating high-level goals into motor commands. The company's decision to build its own hand gives it a consistent hardware platform for data collection. When the model and the hand are designed together, engineers can tune sensors, joint ranges, and control policies in a more integrated way. This full-stack approach is expensive, but it can reduce the friction that occurs when software is trained on one robot and deployed on another.
The robotics industry has seen a wave of humanoid and dexterous-hand projects in recent years. Some companies focus on logistics, others on manufacturing, healthcare, or household tasks. Many rely on simulation, imitation learning, and teleoperation to bootstrap capabilities. Genesis appears to be pursuing a similar playbook, but with an emphasis on human-like hands and wearable data capture. If the model can learn from workers in pharmaceuticals or manufacturing, it could accumulate data in domains where precision and repeatability are highly valued.
Data, Labor, and the Factory Floor
The sensor gloves are a practical attempt to solve the data bottleneck. Instead of asking workers to stop and program a robot, the gloves can record natural hand movements. Those recordings can be paired with video and other sensor streams. Over time, the AI model can learn to reproduce certain actions. In pharmaceuticals, for example, tasks might include handling vials, assembling kits, or performing repetitive laboratory procedures. In manufacturing, tasks might include inserting components, inspecting parts, or operating tools.
However, the same data collection raises questions about consent and job security. Workers may be asked to wear gloves that record their every movement. If the resulting data is used to train robots that later replace them, trust could erode. Some companies might offer incentives, such as safer working conditions or reduced repetitive strain. Others might face resistance from labor unions and regulators. The outcome will depend on how transparently the data is used and whether workers share in the productivity gains.
There is also a technical question about variability. Human workers adapt constantly to small changes in materials, lighting, and tool wear. A robot trained on a fixed dataset may struggle when conditions shift. Foundation models aim to generalize, but they still need broad and diverse data. Genesis may need to collect data across many sites, tasks, and object types to make GENE-26.5 robust. The gloves and videos could help, but scaling that effort is a major operational challenge.
Engineering Challenges Ahead
Dexterous robotic hands are difficult to build. They require compact actuators, durable joints, tactile sensors, and control algorithms that can handle contact. A human-like hand has many degrees of freedom, which increases complexity and cost. It also needs to be robust enough for real work. If a finger is damaged, repairs must be quick and affordable. Power consumption and heat are additional constraints. A hand that performs well in a demo may overheat or wear out in continuous operation.
Software challenges are equally significant. The model must perceive objects, plan actions, and adjust in real time. It must handle uncertainty and recover from mistakes. Safety is paramount when a robot works near humans. The system needs to detect collisions, limit force, and respond to unexpected movements. These requirements are why full-stack integration can be an advantage. Genesis can design the hand and the model together, potentially reducing latency and improving reliability.
The company is also pursuing a full-body, general-purpose robot. That step dramatically expands the scope. A mobile robot must navigate, balance, and manipulate. It must integrate batteries, compute, sensors, and communication. The market for general-purpose robots is still emerging, and customers may initially prefer single-purpose automation. Yet the long-term prize is a machine that can perform many tasks without reprogramming. Genesis is betting that a human-like hand and a strong foundation model will be key components of that future.
The seed funding round was co-led by Eclipse and Khosla Ventures, with additional backing from Bpifrance, HSG, former Google chief Eric Schmidt, telecoms entrepreneur Xavier Niel, MIT Computer Science and Artificial Intelligence Laboratory director Daniela Rus and Apple distinguished scientist Vladlen Koltun. The company said its team of 60 people is split roughly evenly between the US and Europe, with a slightly larger group in the US base in California. It added that it is working to supplement the mechanical hand with a full-body, general purpose robot.
Source: Silicon UK News