Wayve builds self-driving AI that learns like a human

Published: 2026-06-29T12:27:56.000Z

Most self-driving cars behave a little like a pupil who has memorised one particular exam, leaning on detailed maps of every street. Wayve, a British company that grew out of Cambridge, has chosen a different route, building AI that learns from experience and adapts to roads it has never seen before, much as a human driver does.

Most self-driving cars on the road today behave a little like a pupil who has memorised the answers to one particular exam, since they lean on painstakingly detailed three dimensional maps of every street they travel, alongside arrays of costly sensors and reams of hand coded rules. Wayve, a British company that grew out of the University of Cambridge, has chosen a fundamentally different route, and it is one that resembles the way a person genuinely learns to drive. Rather than being instructed what to do at every junction, Wayve's system learns from experience, watching, interpreting and adapting to roads it has never encountered before, which the company contends is the only approach capable of scaling across the untidy variety of the real world. The technical heart of all this is what Wayve calls Embodied AI, or its AI Driver. Founded in 2017 by the machine learning researchers Alex Kendall and Amar Shah, the company does away with high definition maps and lidar in favour of a system that drives chiefly using ordinary cameras and a basic satnav route. It is built upon a neural network, an algorithm loosely modelled on the human brain, that learns patterns from vast quantities of video and driving data. Kendall has put the comparison rather neatly, observing that when a human learns to drive, they arrive with sixteen or seventeen years of spatial awareness and hand to eye coordination already behind them, and Wayve is trying to give its AI a comparable grounding of general understanding rather than a rigid rulebook. The manner in which it learns is the clever bit. Wayve combines imitation learning, where the system copies the behaviour of expert human drivers, with reinforcement learning, where it improves each time a human safety driver has to step in. The reward is generalisation, the prize the whole industry covets, since a car trained in this way can be set down on unfamiliar streets in a different city and still cope. The company has shown its software driving on never before seen roads in Cambridge and threading through the narrow, chaotic streets of European cities, all without any pre mapping. Tellingly, Wayve has no wish to build the entire stack, since it makes neither the cars nor the mobility network, and instead positions its AI Driver as software that partners can embed across many different vehicles. That partnership strategy is now bearing genuine fruit. Wayve has raised 1.5 billion dollars to roll out its platform globally, works closely with Microsoft's Azure cloud to train its models, and has struck a landmark agreement with Uber to launch public road trials of fully autonomous vehicles in the UK, beginning in London before widening to cities such as Tokyo. It has also trialled autonomous grocery deliveries with Asda and attracted investment from Ocado. The argument throughout is scalability, since a mapless, camera based, vehicle agnostic system is considerably cheaper and quicker to deploy than the heavily engineered alternatives. The regional thread leads straight to the Gulf. The UAE has been one of the boldest markets anywhere for autonomous mobility, with Dubai aiming for a quarter of all journeys to be driverless by 2030 and regulators waving through trials that would stall for years elsewhere, while Saudi Arabia weaves autonomy into projects such as NEOM. A system designed to adapt to new cities without months of expensive mapping suits the Gulf's fast changing, newly built road networks rather well, which makes Wayve's approach an unusually natural fit for the region's ambitions.