Machenta raises US$40 million to develop one-demonstration robot learning
The London startup uses simulated data to train its model, but precise tasks and complex movements remain difficult.
A US$40 million seed investment will support Machenta’s development of an AI model designed to teach robots unfamiliar tasks from one demonstration. Imperial College London announced the funding on 27 October, naming Creandum and Union Square Ventures as co-leaders of the round.
The London startup, an Imperial spinout, will put the funding toward hiring and expanding the computing resources and simulated data used for training. A move to new office and laboratory space in King’s Cross, London, is also underway. Imperial called the investment the largest seed financing raised by a UK robotics startup.
Edward Johns, Machenta’s CEO and co-founder and an Imperial professor of robot learning, told Business Insider that training relies entirely on computer-generated environments. He said the trained model can then direct a robot through an unfamiliar task after observing it once, without further training.
According to Imperial, a single model has learned packing and tidying tasks on multiple types of robots, including humanoids. The research started in Johns’s Robot Learning Lab six years ago. He later established Machenta with former PhD student Vitalis Vosylius.
Business Insider reported that the company employs 11 people, is still focused on research and has no near-term plans to sell the technology. Its model continues to face difficulties with precision tasks and complex movements. Machenta did not disclose a valuation.
