Machenta, a London-based robotics startup, has raised $40 million in a seed funding round to advance a contrarian approach to artificial intelligence: teaching robots to perform complex physical tasks without ever using real-world data. The funding round was co-led by venture capital firms Creandum and Union Square Ventures, as the 11-person team attempts to challenge Silicon Valley rivals currently valued in the tens of billions of dollars.
Founded by researchers from Imperial College London, Machenta is betting that synthetic, computer-generated environments can replace the massive amounts of real-world footage and data that other companies are currently scrambling to acquire. Edward Johns, Machenta’s CEO and a professor at Imperial College, stated that the company’s model is trained entirely in virtual spaces that mimic how physical objects move and interact. According to Johns, the system does not utilize a “single pixel of real-world data” during its foundational training phase.
The core of Machenta’s value proposition is its “one-shot” learning capability. Once the AI model is trained in its virtual environment, it can theoretically direct a robot to perform a previously unknown physical task—such as tidying a desk or placing items into a container—after witnessing the action performed just once by a human. This eliminates the traditional need for thousands of hours of repetitive physical training or manual coding for every new motion. In demonstrations, the company has shown its robots successfully packing and sealing boxes on an assembly line after a single visual demonstration.
This simulation-first methodology places Machenta in direct competition with a field of heavily funded U.S. competitors. Among them is California-based Physical Intelligence, which holds a valuation of approximately $11 billion, and FieldAI, which has sought funding at a $10 billion valuation. Another major player, Skild AI, is valued at more than $14 billion and recently reported $100 million in annual recurring revenue from a customer base of over 60 clients. While Skild AI has demonstrated models that can perform tasks lasting up to 10 minutes following a single video demonstration, Machenta’s focus remains on the algorithmic challenge of generating the synthetic data itself.
The startup’s origins trace back six years to Johns’ lab at Imperial College London. Early experiments in robot learning were far more labor-intensive; researchers had to physically guide a robot arm through a task, after which the machine would require roughly 30 minutes of additional data collection regarding its environment before it could repeat the movement independently. Johns and co-founder Vitalis Vosylius, a former Ph.D. student under Johns, spent several years refining the technology to eliminate that half-hour delay. They launched Machenta late last year after determining the model could finally execute tasks immediately following a demonstration.
Despite the significant seed investment, Machenta remains a research-oriented firm with no immediate plans for commercialization. The technology still faces hurdles, particularly regarding high-precision tasks and exceptionally complex movements which remain difficult for the model to master in its current form. The company plans to use the $40 million infusion to expand its engineering team and scale its computing power, with the ultimate goal of refining its simulated environments to support more intricate robotic applications.
