A British start-up, Vsim, has developed an innovative software that allows robots to rapidly learn new tasks through virtual simulations. In a recent demonstration, their robot, Freddo, was able to quickly master the skill of walking, recognising, and grasping a plastic bottle—achieving this within just a few minutes, a process that traditional systems could take days to replicate.
Behind the Innovation
Founders Michelle Lu and Kier Storey established Vsim in Cambridge in 2022 with the vision of creating a high-performance training system for robots. Leveraging their past experience with Nvidia’s Isaac Sim, a similar simulation tool, they optimised their software to fully exploit the potential of advanced graphics processing units (GPUs). This resulted in a simulator that remarkably outpaces others in speed and efficiency.
Vsim’s technology allows Freddo to execute tens of thousands of simulations while on the move, enabling it to anticipate and adapt to dynamic, unstructured environments such as homes. Storey explains this capability helps the robot react safely and effectively to unexpected changes, like interacting with humans or animals.
Industry Comparisons
Despite being a small team with just 10 engineers, Vsim’s rapid simulation capabilities have drawn attention. In contrast, industry giant Nvidia employs hundreds of engineers within its extensive robotics division. Though they lead in AI chips and provide comprehensive robotics software, Nvidia acknowledges that their solutions only provide a basic understanding of the real world, particularly for complex, long-range tasks. Nvidia has thus shifted to using AI agents to build and evaluate virtual training environments more efficiently.
Broader Context and Challenges
Associate Professor Rika Antonova from the University of Cambridge highlights the promise of Vsim’s accelerated simulations, which allow robots to adapt in real time by analysing millions of scenarios instantly. However, she notes that current simulations still struggle with mimicking certain real-world conditions, like the handling of deformable objects. Addressing these limitations is a priority for both Vsim and Nvidia.
To further advance their technology, Vsim plans to introduce a second robot, Nacho, to accompany Freddo in their developmental efforts, enhancing the adaptability of their software across various robotic platforms.
Reporting based on coverage first published by bbc.co.uk.