Table of Contents
The Complexity of Urban Autonomy
Autonomous driving is not just a matter of software; it is a holistic challenge that spans safety, infrastructure, simulation, data, and embedded architecture. The industry has spent the last decade building systems from the ground up, and the stakes are high. The weight of the entire industry rests on how well a company can translate a vision into a reliable, scalable product that can operate safely in the real world.
Business Models in the Autonomous Landscape
Three main approaches dominate the market today. The first is the vehicle‑centric model, exemplified by Tesla, where the company sells its own cars and integrates its autonomous stack directly into the vehicle. This model limits the company to its own brand and fleet. The second is the fleet‑centric model, used by Waymo, which builds and operates its own autonomous fleet, a capital‑intensive operation that expands city by city. The third, and the one chosen by the startup in focus, is a licensing model that offers its embodied AI platform to other fleets and manufacturers. The belief is that most fleets and manufacturers—whether in automotive or other physical industries—will find it more efficient to partner with a scalable AI platform than to build one from scratch.
Sensor Choices and Scalability
Sensor strategy is a key differentiator. Waymo relies heavily on mapping, lidar, radar, and a suite of sensors to create redundancies that enhance safety. Tesla, on the other hand, leans more on cameras and neural networks. The startup’s approach is to build a foundation model that can drive with camera‑only systems, but it also supports radar and lidar for those products that require them. The goal is to deploy autonomy in any vehicle, anywhere, which means the system must be adaptable to different sensor configurations. The result is a low‑cost, on‑board AI stack that uses six cameras and one radar, with compute and sensor costs in the hundreds of dollars—well below the price point of many existing solutions.
Testing in London’s Streets
London’s ancient road network presents a unique set of challenges: far more roadworks, cyclists, pedestrians, and complex roundabout interactions than cities like San Francisco. These conditions forced the team to adopt a scalable approach that does not rely on pre‑built indicator or traffic light detectors. Instead, the AI learns directly from data, building a world model that predicts how the environment will evolve. During a test drive in King’s Cross, the vehicle navigated traffic, diversions, and cyclists using only on‑board intelligence, making decisions in real time without any external guidance. The experience was described as smooth and confident, a testament to the effectiveness of the data‑driven approach.
The Path Forward for Embodied AI
As the startup scales up data and compute by partnering with large fleets worldwide, the AI platform is expected to become even safer and more cost‑effective. The company’s vision is to provide a single, scalable solution that can be integrated into a wide range of vehicles and physical industries beyond cars. By leveraging the collective data from its partners, the platform can continuously improve, offering a performance edge that no single manufacturer can achieve on its own. The future of autonomous driving, according to this perspective, lies in embodied AI platforms that can adapt to diverse sensor setups, operate safely in complex urban environments, and scale globally through licensing rather than proprietary vehicle sales.