Humanoid robots are no longer just things you read about in science fiction.
An increasing number of companies in Europe are building robots to aid humans in everything from manufacturing to everyday domestic tasks — using AI to help them “see”, learn how to walk and manipulate objects.
The latest robotics company to enter the fray is ETH Zurich spinout Mimic, which has just raised $2.5m for its AI-powered robotic hands which are designed to ‘mimic’ real human movements.
The round is led by early-stage Swiss investor Founderful, with participation from another.vc and Tiny VC.
Investor interest in robotics is at an all-time high, especially amid the AI frenzy. Goldman Sachs forecasts the global market for humanoid robots could reach $38 billion by 2035.
AI-driven robots
Mimic is among the companies that are applying recent advances in large language models (LLMs) — the same type of AI system that powers ChatGPT4 — to help robots “understand common sense and the physical world,” as Mimic cofounder Elvis Nava puts it.
The biggest players working on humanoids are based in the US — such as Figure, which raised $75.5m in February this year for its full-bodied humanoid robot driven by generative AI.
But European robotics companies are on the rise too. In January this year, Norwegian AI startup 1X raised $100m for its AI-powered two-legged robot butler and ETH spinoff ANYbotics raised $50m in May 2023 for its four-legged autonomous robots for use in industrial applications.
Mimic’s edge over its competitors, it says, is that it’s not trying to build an entire robot, but focuses on manipulation with hands. Its robotic limb is also not designed for use in a specific industry, but is “general purpose”, able to do a variety of tasks — be it recycling or machine tending — with a single robot design.
Mimic’s foundational AI model is trained on large-scale data recordings of humans performing manual labour tasks, which enables its robot to imitate human movements.
One of the imagined use cases for this technology is in an industrial bakery. Bread is still sorted and packaged largely by humans, says Nava, but Mimic’s robot could take over these tasks for them.
Another use case is in the pharmaceutical industry. Diagnostics machines are typically steered by highly skilled technicians, who have to pick up vials, pipette samples into them and then insert the vials into the diagnostics machine — something that Mimic says its robots could do instead.
The beauty of LLMs, says Nava, is that they can easily be integrated into robots to allow you to talk to them.
Our vision is to have a general purpose foundational model so that you can give the robot language commands, like ‘Robot, pour me a beer.
“Our vision is to have a general purpose foundational model so that you can give the robot language commands, like ‘Robot, pour me a beer”, he says.
He adds that applying language and visual understanding to robots is easy, but getting them “proficient in motion” remains the hardest part.
Challenges ahead
Mimic has five full-time employees and is working towards officially launching a product later this year for commercial clients.
One of the challenges Mimic will face is scaling up its hardware. While it’s easy to set up a pilot project and have the robot work for one customer, doing that for hundreds of customers is another thing entirely, says Nava.
What helps is that the company doesn’t produce the humanoid arm of the robot itself, as other humanoid companies do.
It buys the robotic arms which are already widely sold and creates its own hand to attach them to, which will help the company “scale faster,” says Nava.
Additionally, Mimic is constrained by the limited availability of robot training data, used to teach robots how to move and interact with the physical world. It's a problem that many robotics companies worldwide are coming up against, and is one of the main things holding progress in robotics back.
However, Mimic's approach of using video data of humans doing tasks — of which there is a lot out there — to train its model eases this problem.
Strengths of ETH Zurich
One of Mimic’s strengths is that it is part of a supportive ecosystem in Zurich, largely shaped by ETH, which is the most successful university in Europe in terms of spinning out companies. The university, which has its own centre for robotics, currently has 26 companies working on robotics, though only a few are working on aspects of humanoid technology, according to its data.
“ETH is uniquely good in Europe for having a great pool of talent — especially in robotics. Usually it is hard to find talent outside the hubs of the US or the UK,” says Nava.
“It’s one of the few places in continental Europe where you can spin out a company and have a chance of it succeeding,” he adds. “(The terms) are not perfect, but it’s better than the deal that most universities can give you.”
