Wayve

The road to autonomous rides in Japan is officially paved! 🇯🇵 We’re teaming up with Hinomaru Kotsu, Nissan Motor Corporation, and Wayve to bring robotaxis to Tokyo starting in late 2026. We can't wait to deliver the next generation of everyday transportation to riders in Japan.

Wayve’s Vice President of Engineering, Brandon Basso, joined the The Driverless Digest podcast to discuss GAIA-4 and how generative world models are opening up new possibilities for autonomous driving development, safety and validation. The conversation dives into what makes validating an end-to-end AV2.0 stack uniquely challenging, how closed-loop simulation can help us capture complex driving scenarios, and why GAIA-4’s ability to incorporate radar alongside vision is so important. Thanks to Ben Hubbard for the conversation on how richer, more interactive world models can help us us test and understand the behavior of our driving technology across diverse scenarios and markets, complementing real-world testing as we continue to advance embodied AI for autonomous driving. Listen to the full episode on Driverless Digest: https://lnkd.in/eX3dDyqS

I'll be in SF next week speaking at #Actuate26 with Adrian Macneil. Lots to debate in robotics today! I'll share Wayve's latest contrarian ideas: how we are building a frontier ecosystem for Embodied AI.

GAIA-4 paves the path to a new way to validate autonomy. The shift to end-to-end driving models is an industry-wide one: any team credibly building end-to-end autonomy inherits a system that can no longer be taken apart and tested in pieces, and the simulation tools built for modular stacks cannot validate it.  GAIA-4 turns an end-to-end AI Driver from a supposed “black box” into a system that can be examined, validated and trusted. Every intervention can be replayed and interrogated, every model change measured against the same real scenarios, and every sensor the vehicle depends on can be put under test.  This introspection, repeatability, and evidence are the foundation of a new validation playbook, Simulation 2.0, enabling us to scale safely across millions of vehicles worldwide with our automaker partners. Read the full blog: https://lnkd.in/eSyJw4CV

Does radar change the AI Driver’s behaviour? Useful simulation for autonomy needs to go beyond camera realism. GAIA-4 extends this capability with multimodal generation, bringing radar into the simulation alongside cameras. Radar is particularly important for redundancy and safety-critical sensing, providing robustness in adverse conditions such as fog, spray, and low light, where visibility is poor.  With this multimodality established, we can start to understand radar’s impact on model performance by simulating scenarios with and without it. In most cases,  the addition of radar measurably shifts the closed-loop trajectory, providing evidence that the generated radar carries decision-relevant information and ensures the vehicle decelerates appropriately. https://lnkd.in/eSyJw4CV

How do you know what the AI Driver would have done? How does a new model behave in exactly the situation that tripped up the last one? With GAIA-4, we can ask these kinds of safety-critical questions through counterfactual simulation. We can introspect and debug the AI Driver by taking any real moment, changing one thing, and watching what the AI Driver would have done in a fully closed loop environment. This enables us to better understand the model’s behavior and turn an on-road event into a repeatable diagnosis of the model's behavior. This creates a validation flywheel where validation stops scaling with road miles and starts scaling with compute: efficient enough to run on every candidate model, fast enough to move deployment forward and focused on the eventful miles that matter most in model validation. https://lnkd.in/eSyJw4CV

Big news for autonomous mobility in London 🇬🇧 Transport for London has granted Private Hire Vehicle licences to Wayve's autonomous vehicles, an important step forward for our partnership with Uber to bring supervised autonomous rides to the capital. Since June, more than 100,000 Londoners have already signed up to be among the first riders. Later this summer, a select group who joined this interest list will get the chance to take an early trip as we fine-tune the experience ahead of full public launch. Built and trained on UK roads, the Wayve AI Driver has been learning in London since 2018. This milestone reflects years of collaboration between Wayve, Uber, regulators, and local authorities to help bring autonomous rides to the capital safely and responsibly, and we're exited for what's to come.  https://lnkd.in/eETqFqCE

How do you know the simulation is trustworthy? Before we can trust the AI Driver, we have to trust the world around it. In GAIA-4, every vehicle, pedestrian and cyclist continues exactly as they did in the original recording. Only the AI Driver changes. We call this World-on-Rails, a conservative, repeatable baseline for safety validation. That constraint gives the simulation two properties essential for validation. First, other road users stay grounded in real behavior. Second, the evaluation stays conservative: no vehicle, pedestrian, or cyclist changes its behavior, preventing the simulator from altering a safety-critical outcome. Under the world-on-rails constraint, we can assess the safety outcome for any scenario, including those where the vehicle safety operator intervened. https://lnkd.in/eSyJw4CV

One of the biggest questions about end-to-end AI isn’t whether it can drive. It’s whether it can be proven safe. The very thing that makes these models powerful - that they operate as one learned system rather than a collection of separable parts - breaks many of the tools the industry has historically used to prove safety. Including simulation. That is why end-to-end AI needs a new validation playbook. We call it Simulation 2.0. And GAIA-4 is one of its foundational building blocks. Building on our pioneering history of world models, GAIA-4 puts the AI Driver in control and generates the future its decisions create. This lets us move beyond evaluating what the AI Driver predicts and allows us to understand what it would actually do, answering a new set of questions about the most safety-critical scenarios. Read the full blog: https://lnkd.in/eSyJw4CV

Alex Kendall shares his thoughts on why autonomy and embodied AI ecosystems can’t be boiled down to a simple build vs. buy decision.