Eyes-Off Autonomous Driving Arrives by 2028 | Brandon Basso

Wayve chases eyes-off driving by proving end-to-end safety

Wayve is betting that self-driving arrives as an end-to-end learned system that can be validated to a standard human drivers never meet. The company builds one AI driver for all vehicles rather than a single carmaker's fleet, and starts with hands-off level 2 plus before the eyes-off level 3 systems Basso expects in 2027 and 2028. Its harder problem is proof. Wayve treats validation as a data problem, leaning on a global data operation and a learned safety representation instead of map-based checkers that cap performance.

Takeaways

• Eyes-off driving arrives in 2028. Basso predicts the industry shifts in 2027 into 2028 from level 2 plus hands-off assistance to level 3 eyes-off systems. The car takes responsibility for safety while the driver checks out and does other tasks.
• Wayve targets all vehicles, not just Teslas. Wayve's mission is ubiquitous autonomy for everyone, starting with hands-off level 2 plus driving in urban and highway settings and reaching a level 4 robotaxi, for which it runs a supervised public pilot in London.
• Validation is the unsolved half of autonomy. Basso splits the problem into making an AI driver better than a human and proving it works, and he works on proof. There has been little rigorous thinking on validating end-to-end systems, the same problem frontier AI labs face.
• Safety has no single metric. Wayve breaks safety into safety, comfort, compliance and competence, and aligns its metrics with human expectation because there is no leaderboard equivalent. Traffic behaviour differs between the US, Germany, Japan and the UK.
• There is no general dataset for driving. Physical AI has no shared corpus like language does, so Wayve treats both driving and validation as data problems. It started that work a decade ago and secured geographic and vehicle diversity early.
• Checker systems become the bottleneck. Wayve uses a learned occupancy map marking where it is unsafe to drive, guiding trajectories. Modular systems lean on high-definition maps and checkers, which break when a stop sign is missing or has moved, capping system performance.
• Crash deaths hit young, new drivers. Fatalities fall disproportionately on younger, less experienced drivers. Wayve is gathering evidence that a level 2 plus hands-off driver beats expert humans, while level 3 eyes-off needs guarantees close to robotaxi standard, giving people their commute time back.

How Wayve builds its data

• First-party collection, run by a global operations team working 24/7, 365 days a year.
• Purchased and licensed data, including dashcam data going back ten years.