Our road ahead

In 2017 everyone believed autonomous vehicles were here. Billions of dollars were invested. Partnerships were signed. There was a gold rush to expand DARPA-era self-driving technology to different business models: manufacturing, licensing, trucking, delivery, and robotaxi.

You would have been crazy to found a contrarian start-up to compete in this space against trillion dollar technology giants. Well, maybe we were.

At the same time as the autonomous driving industry was scaling this traditional robotics stack—which typically relies on an expensive array of sensors and is operationally limited by HD maps and rules-based control strategies—a quiet revolution was happening in machine learning.

From an early age, I have always had a fascination with creating machines which can benefit our world. I believe there is no greater challenge than building machines that can see and act. But today, AI only exists in the virtual world through software and apps, despite the well-known fact, “We have a brain for one reason…to produce adaptable and complex movements.” It was clear to me that the next transformative step was to build embodied AI which could physically interact with our world.

I was fortunate enough at this time to be studying computer vision and deep learning for my PhD at the University of Cambridge, with a supervisor and an environment which encouraged entrepreneurship. Our award-winning research showed for the first time that it was possible to use deep learning to teach a machine to understand where it is and what’s around it, plus crucially understand what it doesn’t know. With this breakthrough technology, I imagined we could now move away from what we predicted to be the prohibitive hurdles to scale such as HD-maps, lidar and rules-based autonomy, towards machines that have the intelligence to make their own decisions based on what they see with computer vision.

We set out to develop AV2.0: the next generation of autonomous driving

I firmly believe that to have impact you need to take a bet on what will be possible in the future, not rely on what is possible today. In 2017, machine learning approaches to robotics were limited to experiments in simulation where reinforcement learning was only able to learn behaviours over millions of simulated experiences. But, the results were remarkable and able to solve some astounding problems like beating the world champion of the game of Go. Further advances in machine learning—such as self-supervised learning in computer vision, new techniques for probabilistic and generative modelling, and model-based reinforcement learning for control—were producing opportunities to create intelligent machines that can interact openly with society, and with limited human supervision. The momentum was building and I was convinced that the future of intelligent robotics will be driven by machine learning. Everything we’ve learned at Wayve since then has reinforced this belief.

Wayve began in a garage, developing a Driving Intelligence based on these ideas. The results were quickly promising. For the first time in the world, we showed a reinforcement learning system learning to drive a real-life autonomous vehicle from computer vision. In our first year we demonstrated model-free and model-based reinforcement learning driving our car, sim2real and more.

In five quick years we established the necessary ingredients to pioneer AV2.0: data, compute, partnerships, operations, and—most importantly—our team. We’ve solved the fundamental technical challenges and earned the right to build AV2.0 at scale.

Our vision—autonomy that propels the world forward—will launch society into the age of embodied intelligence. For the first time, we will see machines we trust physically interacting with our world, enriching people’s lives, freeing them to focus on what matters most. This will mark the beginning of a new era of technology that promises to be more transformative than any that came before it. We are excited to be pioneering the embodied intelligence to build autonomous technology that can adapt to the needs of people worldwide—starting with driving on roads.

The road ahead

Our Series B is an inflection point. We’ve partnered with leading global technology investors to raise a $200m Series B, bringing our total capital raised to $258m. Our full-stack team of 140 people has proven leadership across all areas — from hardware to software to autonomy to product. Our fleet is already testing UK-wide, collecting massive scale training data through our commercial fleet partners. We’ll shortly be launching our last-mile grocery delivery pilots, ensuring we have the necessary feedback and environment to develop autonomy with strong product market fit. It is incredibly exciting to see these resources translate into a continuously improving autonomous vehicle, learning from experience, with the data showing that we’re accelerating towards a level four (L4) autonomous future.

Being headquartered in London has given us the fresh perspective of an international city to build a differentiated approach to push a global category. The UK’s heritage in AI is unparalleled, arguably having invented the subject through the work of Alan Turing. But we are building a global product with a global team and with this capital are now positioned to compete on the world stage. Our team is spread across several countries and we aim to be the first company to deploy autonomy in 100 cities.

Although we will use this capital to scale our company, we firmly believe in the core cultural principles our team has embodied since the beginning. Our culture focuses on paving new roads and exploring unknown horizons. Our diversity is our strength. We embrace growth mindsets and are innately curious to learn. Just like our autonomous vehicles learn through fleet learning, we also strive to iterate quickly with clear and effective feedback.

I’m excited to now look ahead towards solving the grand challenges for learned driving. These include amassing large scale off-policy datasets for continual fleet learning, understanding and refining our data curriculum, optimising our Driving Intelligence’s reward function and achieving new levels of scale—with billions of parameters and petabytes of data. The list goes on, simulating the complexity of the real-world, architecting the physical hardware system. What we are building at Wayve is one of the greatest technical challenges of our time. We’re on our way, and hope to bring together people who have a passionate desire to tackle the hardest problems and want to change the world with us.