In 2019, the autonomous vehicle (AV) industry faced a rude awakening. Up until this point, seemingly endless amounts of venture and corporate capital had poured into the industry, thanks to boundless excitement about the potential of the technology. Yet, cracks in the foundation — high profile legal cases and high-stakes accidents — began to emerge. Suddenly, the foregone conclusion of achieving fully autonomous driving had become an open question, and the tremendous personal, economic, and societal benefits seemed uncertain at best. Even before the AV narrative was derailed by these setbacks, I had my own doubts about the prevailing approach to this complex autonomy problem. I did not believe the traditional robotic stack — essentially the same one being used in much simpler application environments like supply chain warehouses and manufacturing, where we had to overcome significant engineering and operational challenges to achieve the necessary reliability and safety metrics — could safely transport humans in self-driving cars on the road at scale.

Around this time, I met Alex Kendall, Co-Founder and CEO of Wayve, a company reimagining autonomous mobility through embodied intelligence. Alex founded Wayve in 2017 after completing his PhD as a Woolf Fisher Scholar at Cambridge, where his research focused on machine learning (ML). We bonded over our shared conviction that AV1.0 was fundamentally flawed — companies had spent billions of dollars and yet, still had a long way to go in order to make their businesses make sense economically. Observing the remarkable results from ML research projects like AlphaGo and the rate at which fields like computer vision were advancing, Alex and his team concluded that large models, trained on significant amounts of data, could be the pathway to developing a next generation autonomous system able to quickly and safely adapt to new driving domains anywhere in the world. 

After years of extensive research and development, Wayve was the first company to apply Embodied AI to develop and test an end-to-end (e2e) deep learning autonomous driving system on public roads. Since Eclipse led the company’s Series A in 2019 and Series B in 2022, Wayve has built the foundation models for autonomy, similar to ‘GPT for driving,’ that can empower any vehicle to see, think, and drive through any environment. Today, the team announced its $1.05B Series C led by SoftBank, with participation from new investor Nvidia and existing investor Microsoft to propel its mission forward.

Over the last 20 years, companies that have adopted the AV1.0 approach have relied on a broad set of predefined human-made concepts, paired with logical/mathematical models that attempt to translate these concepts into optimal driving decisions. Though these concepts and models are designed to cover a set of known, predictable scenarios, they are incapable of generalizing to unexpected situations outside their training data. As drivers, we all remember encountering curveballs we didn’t practice in driver’s ed: an animal running into the middle of the road, debris flying at the windshield out of nowhere, and so on. While rare, these scenarios demand careful consideration to ensure the utility and safety of driving systems. Unlike humans, whose understanding of the world enables us to adapt to unpredictable situations with relative ease, traditional AV technology has not been able to consistently handle these outliers. In other words, AV1.0 is too rigid. 

In contrast, Wayve's approach — AV2.0 — uses self-supervised learning (SSL) techniques to process raw sensor data from vehicle sensors and directly outputs driving commands into driving trajectories, bypassing traditional rule-based systems. This method leverages vast amounts of raw, unlabeled data, allowing the AI to learn and adapt without explicit human annotation. The AI model's effectiveness grows with the diversity and volume of data it processes, becoming more nuanced and adept at driving tasks not just specific to its training environment, but beyond new, unseen environments. Training on large-scale driving datasets without human intervention underscores the robust and transformative impact of SSL in autonomous driving. 

While there is significant differentiation behind Wayve’s model, scaling embodied AI in safety critical applications requires innovation across both a broad AND deep technical stack — reliable hardware development, operationalizing a fleet for data collection, data management infrastructure, scaled training capabilities, robust evaluation, simulation, and safety certification. By innovating in each area, Wayve has been able to create a 10x improvement in driving performance over the past year, while spending an order of magnitude less capital expended than their peers in the AV industry. 

Wayve has established itself as a world-class leader in Embodied AI research. A brilliant technologist, Alex has the rare combination of strong leadership skills, ability to build from 0-1, and parsing complex technical details. Since the first day we met, he was ready and willing to do whatever it takes to build an enduring company. He's assembled a team of top experts in their fields, including Chief Scientist Jamie Shotton, VP of Hardware Dan McCloskey, VP of Commercial and Fleet Operations Kaity Fischer, VP of People Emma Baillie, VP of Software Silvius Rus, and most recently President Erez Dagan. Together, they've pushed the field of autonomous vehicles forward across many dimensions:

- AI research: Wayve's novel multimodal and generative models, including LINGO and GAIA, are advancing AI capabilities in vehicles for intuition, language-responsive interfaces, personalized driving, and co-piloting enhancements.

- Simulations: The team's pioneering work with dynamic neural rendering techniques NERF’s promises to dramatically reduce the need for real-world data, which would reduce the cost and accelerate the velocity of performance improvement.

- Critical Safety Applications: The application of AI in driving demands more than just adding a large model to a vehicle; it requires specialized model architectures, extensive fleet learning, diverse driving datasets, and access to supercomputing. It requires smart regulations that both protect consumers and foster innovation.

Wayve’s team has dedicated years of focused development to gain a competitive edge in these key areas, which uniquely positions them to unlock the vast potential of Embodied AI for automotive and beyond.

This investment will enable Wayve to fully develop and launch the first Embodied AI products for production vehicles. The company will also focus on scaling its foundation models, advancing Embodied AI research, and building an industry-leading AV2.0 Platform with reliable simulation, measurement, and active learning tools for automotive applications. With this funding, Wayve plans to expand operations and partnerships in new markets, building geographically diverse data assets and attracting global talent. 

Driving Towards Prosperity

Since the company’s inception, Wayve has led AI innovation, pioneering the once-contrarian concept of end-to-end AI for the physical world. Wayve is now reshaping the entire industry through its visionary approach. The team’s commitment to challenging the status quo and their persistence in seeing it through demonstrates their capacity to make a meaningful, long-lasting impact on the world. Transportation is a cornerstone of economic productivity. Wayve’s platform promises to improve how people and goods move through the world. The company has all the ingredients to realize their vision of reimagining autonomous mobility through embodied AI, and beyond. 

Congratulations to Alex and team!

Follow Eclipse on LinkedIn for the latest on the Industrial Evolution.