Machenta Raises $40M to Develop Robots That Learn From One Demo

LONDON -- British robotics startup Machenta has raised $40 million in seed funding towards building a robotics artificial intelligence model that will enable robots to learn new physical tasks through just one demonstration by a human.

The startup announced the fundraising Oct. 27, with funding led by venture firms Creandum and Union Square Ventures. With the funding, Machenta aims to hire more talent and develop the robotics foundation model within its lab and office in the King's Cross area of London.

Unlike typical robot learning methods that use huge amounts of training data based on the real world, Machenta claims its model uses pretraining within simulation and no real-world images.

The model enables a person to demonstrate a task only once, and then enables the robot to learn to perform the same task right away without needing to provide further examples or train the model to carry out the specific task.

The startup has released videos showing robots carrying out tasks like sorting and organizing items after being shown a demonstration once. According to Machenta, the robots had not been introduced to these items during training.
Nevertheless, the company has not reported how consistent these technologies are in varying settings or with more complex industrial applications. There have been no reports about any commercial deployment.
The technology was invented by Edward Johns, who is a professor of robot learning at Imperial College London, and by his former PhD student, Vitalis Vosylius. The technology is based on six years of work at the Robot Learning Lab at Imperial College London.

The company plans to spend its money on scaling the technology and growing its team of researchers. There is no information about its commercial release and clients.