When it comes to self-driving cars, the same picture usually pops into your head: a car decked out with lidars, sensors, and cameras, creeping slowly down the street while relying on pre-made maps accurate to the centimeter. This is the approach major companies like Waymo and Cruise have been developing for years — reliable, but very expensive and rigidly tied to specific cities.
British startup Wayve decided to take a different path. Instead of teaching the car to follow rules and maps, they are teaching it to drive — much like how humans learn to drive.
Learning Through Observation, Not Rules
Wayve builds its system on deep learning. Simply put, their neural network looks at the road through cameras and learns to make decisions — where to turn, when to brake, how to steer around an obstacle — based on a huge number of real-life examples.
This is a fundamentally different approach. Instead of programming every rule manually («if you see a pedestrian, stop», «if the light is red, stay put»), the system finds patterns in the data itself. It observes how human drivers behave in different situations and tries to reproduce that behavior.
It sounds simple, but in practice, it requires colossal computing power and a lot of data. And that is where Microsoft Azure comes into play.
Why the Cloud Is Not Just a Convenience, But a Necessity
Wayve uses the Azure cloud platform to train its models. This isn't a technical detail — it's the foundation of the entire approach.
Training neural networks for autopilot isn't a one-time task. The model needs to be constantly improved: adding new data, fine-tuning on complex cases, testing different architectures. This requires powerful GPU clusters capable of processing terabytes of video from cameras and simulations.
Buying and maintaining such hardware in-house is expensive and slow. The cloud makes it possible to scale quickly: if you need more power, you get it in minutes, not months.
Furthermore, Azure provides tools for working with data and models: from storage systems to specialized machine learning frameworks. This allows the Wayve team to focus on algorithms rather than infrastructure.
What Does This Mean in Practice? 🚗
The main advantage of Wayve's approach is flexibility. Their system isn't tied to a specific city or type of road. It learns general driving principles, which means it can adapt to new conditions faster than systems built on rigid rules and detailed maps.
This is especially important for Europe and other regions where streets are narrow, roads are winding, and lane markings are often inconsistent. Traditional autopilots feel unsure in such conditions — they need to re-map every street. Wayve, in theory, can handle this more easily.
Another plus is cost. Wayve's system relies mainly on cameras rather than expensive lidars. This makes it potentially cheaper to manufacture and easier to scale.
But There Are Questions
For all the appeal of the approach, there are a few important points that remain unclear.
First is safety. Neural networks handle typical scenarios perfectly but can behave unpredictably in rare, non-standard situations. How will the system react to something it has never seen in the training data? This is a question that worries not only regulators but the developers themselves.
Second is explainability. When the autopilot makes a decision, it's often impossible to understand exactly why it did so. This complicates debugging and raises questions regarding liability: who is to blame if the system makes a mistake?
Third is data. The more examples the neural network sees, the better it works. But collecting, labeling, and storing such volumes of data is a massive task in itself, requiring not just technology but a clear understanding of privacy and ethics.
Why Microsoft Is Investing in Wayve
In 2024, Microsoft invested in Wayve as part of a $1 billion funding round. This isn't just an investment in a promising startup — it's a bet on the future of autonomous transport built on AI.
For Microsoft, this is an opportunity to show how Azure can become the foundation for complex, resource-intensive systems. Wayve isn't just a client; it's a case study demonstrating that cloud platforms can handle tasks that seemed like science fiction just recently.
Moreover, it's a strategic move. The automotive industry is transforming rapidly, and whoever is first to offer a scalable, affordable solution for autopilot will capture a huge market. Wayve could become that player — if their approach proves its reliability.
What's Next?
Wayve continues to test its system on the roads of London and other cities. They are growing their vehicle fleet, collecting data, and improving models. In parallel, negotiations with automakers are underway — the goal is to integrate their technology into production cars.
It's too early to say whether Wayve's approach will become dominant. But the very fact that someone is seriously challenging the established paradigm deserves attention. It's a reminder that in the world of technology, the most obvious path isn't always the winning one.
Deep learning is transforming many industries — from medicine to logistics. Self-driving cars may become the next area where neural networks prove they can not only help humans but replace complex systems built on rules.
But mass adoption is still a long way off. Ahead lie years of testing, regulatory approvals, and technological improvements. And it's still unclear who will turn out to be right — those betting on precision and control, or those who believe in the ability of neural networks to learn from examples.