Read now
AI in the digital world is progressing rapidly, but intelligence in the physical environment remains brittle. Robots today still struggle with the kind of generalised reasoning and adaptability that would allow them to operate beyond narrow, pre-programmed tasks. This is one of the most important opportunities in applied AI, and it's why we've invested in Base8.
A robot that can pick apples cannot pick tomatoes. A system trained in one warehouse cannot adapt to a new layout. Even autonomous vehicles — perhaps the most advanced robots we have today — struggle to improvise when conditions change, as was the case during Waymo's disruptions in a San Francisco blackout.
Today's robots are remarkable in many ways, but there's still a lot to build when it comes to general intelligence. This is the core problem Base8 is tackling.
Base8 is a micro research lab building a brain-inspired architecture for robot intelligence. The founding thesis draws on how the brain learns: using techniques that will enable robots to not only operate in known situations but also improvise when the environment changes.
There's a wealth of research in this area already, and Base8 plans to marry the best of academia with practical, real-world approaches, starting with the milestone of achieving the equivalent of mouse-level intelligence. This will involve continuous learning and adaptability that reduces the need to retrain a robot from scratch.
Base8's long-term ambition is to become a foundational intelligence layer for robotics. The vision is bold and the technical roadmap is hard, but the market opportunity is substantial. This is precisely the kind of ambition we love to back at the earliest stages of company building.
Two things drove our conviction: an exceptional founder and a strong fit with our economic empowerment theme.
Andrew Hart is a second-time entrepreneur with deep tech roots. At his previous company, Hyper, he led the development of the world's most accurate indoor navigation system. It combined WiFi positioning with SLAM (Simultaneous Localisation and Mapping) to achieve 1-metre accuracy, outperforming big-tech alternatives. Before Hyper, he built a popular open-source project for Apple's AR and Location frameworks that got traction with engineers at Apple and Google.
When Andrew first told us about his new venture, he'd already travelled to the USA multiple times to meet with robotics teams at leading companies. He didn't want to build in isolation. He wanted to see up close what state-of-the-art looked like and what it would take to build a challenger business in the UK. This demonstrated a proactive, intellectually curious, and feedback-seeking orientation that we look for in founders.
Andrew also had a sense of urgency about the opportunity that's infectious. Before raising capital, he was already running experiments and building in public. One example of this is HelloRL, an open-source Python library that simplifies Reinforcement Learning.
Base8 sits within our economic empowerment theme. We believe that robots have the potential to generate positive economic and social impact. They can do work that would otherwise put human lives at risk (e.g. navigating dangerous environments, handling hazardous waste) and also hold promise in fulfilling roles with chronic staff shortages (e.g. in agriculture, construction, or social care.)
Large language models and AI agents have transformed work in the digital realm. The next frontier is the physical world, but that requires going beyond current approaches. Today's Vision-Language-Action models still need huge amounts of training data for narrow tasks, and struggle to generalise when the environment changes. Base8 is building towards a fundamentally different architecture to solve this, and we're thrilled to partner with the company at inception.
"Years back at an AI dinner I hosted, Andrew shared what was then a controversial opinion: generative AI and the transformer architecture are a dead end for general intelligence. He wasn't the only one in the community with this view, but in the years that followed, he's one of a few that have committed to doing something about it. We're excited to join Andrew as his first investor and look forward to supporting both the research and commercial application phases of the business."
— Michael Tefula, Principal and Head of Product, Ada Ventures