Technology companies succeed in part because of good timing. By itself, a cool technology product doesn’t guarantee widespread traction. Designing the “right” solution for the customer — one that can consistently do reliable and valuable work — is paramount.
My time at General Electric building autonomous systems, and the last five years I’ve spent investing in robotics companies at Eclipse Ventures alongside colleagues from Tesla, Flex and SunPower, has taught me that people don’t really care about the bot. They care about reliably getting work done with greater efficiency.
The Next Phase of Robotics
A classical approach to robotics engineering struggles to achieve reliability in complex, unconstrained environments. The current state of autonomous-vehicle development is a perfect and all-too-timely example. Expensive sensors, labor-intensive HD maps, hand-coded rules — this technology stack seems increasingly cumbersome, doesn’t it?
More money, more data, more engineering, and more cars in the test fleet will not solve the problem. We at Eclipse felt that a different approach, a fundamentally new system design, would be needed to bring the elusive autonomous vehicle to the masses.
A New Paradigm for Autonomous Systems
Going back to the initial point, recent advances in technology now make the timing right to tackle the problem from a completely different angle:
- Computer vision has progressed rapidly in the last decade, and an optics-only approach is now sufficient to enable autonomous systems to operate.
- Robust simulation environments make it possible to train machine-learning (ML) algorithms with much less real-world data.
- New compute systems designed specifically for ML training and edge inference will radically reduce the time and resources needed to refine and deploy models that are orders of magnitude more complex than the ones used today.
- “Imitation learning” and “reinforcement learning,” subcategories of machine learning that have long been criticized for being too data hungry and compute intensive, are now becoming viable for applied robotics.
Together, these technological advances now make it possible to build a system that can be taught to operate more reliably than a human. Moreover, they make it possible to build a system at a cost that is economical and won’t require operating expenses to scale linearly with fleet growth — because, really, who wants to spend $180K on a family sedan?
The Evolution of Intelligent Machines
That’s why we’re invested in Wayve, a UK-based company pioneering AI software for self-driving cars. Led by Amar Shah and Alex Kendall, experts in the areas of deep reinforcement learning and computer vision, Wayve is leveraging all of these advancements. Their unique end-to-end machine learning approach allows cars to drive in complex urban environments.
The just–announced $20 million Series A funding round that Eclipse led is allowing Wayve to launch a pilot fleet of autonomous Jaguar I-PACEs in central London — yes, one of Earth’s busiest cities. Participating in the round were Balderton Capital, existing investors, and several undisclosed, preeminent leaders in machine learning and robotics.
See for yourself how, last spring, Wayve demonstrated a self-driving car navigating on roads the vehicle had never been on before. It was accomplished by using only basic cameras, a simple sat-nav route map, and Wayve’s ML “driving brain.”
Video courtesy of Wayve
We are entering a new paradigm of autonomous-system development, one that continues to hold reliability and performance paramount, but dramatically expands the scope of what’s possible for these systems to do in the physical world. And at Eclipse, we are excited to work with early-stage companies that are making this new paradigm a reality.
That’s the common thread across our portfolio of companies: We work with bold, creative entrepreneurs who are disrupting the complex, legacy industries that anchor the world’s economies and who are building the digital bridges into tomorrow.