British start-up Vsim is training robots in virtual environments, allowing them to practise tasks millions of times before transferring the best solutions to physical machines. Its robot Freddo learnt to walk, identify a bottle and grasp it within minutes, demonstrating how faster simulation could help robots perform useful tasks safely in homes and workplaces.

Vsim’s software can run tens of thousands of possible scenarios on the robot itself, allowing it to anticipate events and adapt to people, animals or other machines. The ten-person company is competing with Nvidia, whose larger robotics division offers simulation tools and the Cosmos world model, although even powerful systems still struggle with complex, multi-stage tasks such as filling and pouring from a bottle.