Smart
Our vehicles reason through novel situations in real time, not just ones they've seen in training data. Where rote pattern-matching falls short, we’ve taken a smarter approach to driving.
Our Technology
May Mobility’s autonomy system fuses a predictive world model with a real-time reasoning engine, so our vehicles understand the scene in front of them and think through it — rather than simply matching it against training data. It’s a fundamentally different architecture than the modular AV stacks or pure end-to-end models that dominate the industry, and it’s already available to the public.
Solving the hardest problem in autonomy: handling unexpected scenarios in real-time
Conventional autonomous systems can capably handle situations they’ve trained on. But they struggle outside that training data, and the industry’s answer has been to collect more of it — at enormous cost in time, miles, and dollars.
Our approach is different. Instead of betting everything on having seen every possible scenario in advance, our vehicles reason through the scene in front of them in real time, refreshing their understanding in a fraction of a second. The result is an autonomy system that generalizes to new geographies, new road types, and complex urban environments without the billions of miles and dollars other approaches require.
Solving the hardest problem in autonomy: handling unexpected scenarios in real-time
Conventional autonomous systems can capably handle situations they’ve trained on. But they struggle outside that training data, and the industry’s answer has been to collect more of it — at enormous cost in time, miles, and dollars.
Our approach is different. Instead of betting everything on having seen every possible scenario in advance, our vehicles reason through the scene in front of them in real time, refreshing their understanding in a fraction of a second. The result is an autonomy system that generalizes to new geographies, new road types, and complex urban environments without the billions of miles and dollars other approaches require.
World Model
Our integrated world model develops a detailed understanding of what’s happening on the road in the moment, distilled from physics, the rules of the road, and the unwritten conventions of how pedestrians and drivers actually interact.
We apply that model again and again to imagine what could happen next. Every 200 milliseconds, the vehicle simulates hundreds of possible futures up to 10 seconds out, predicting how each driver, pedestrian, and cyclist’s behavior would shape what everyone else does. The vehicle isn’t matching the moment to a database — it’s thinking it through.
World Model
Our integrated world model develops a detailed understanding of what’s happening on the road in the moment, distilled from physics, the rules of the road, and the unwritten conventions of how pedestrians and drivers actually interact.
We apply that model again and again to imagine what could happen next. Every 200 milliseconds, the vehicle simulates hundreds of possible futures up to 10 seconds out, predicting how each driver, pedestrian, and cyclist’s behavior would shape what everyone else does. The vehicle isn’t matching the moment to a database — it’s thinking it through.
Reasoning Engine
Most AV systems output a single driving strategy and trust it. Ours doesn’t.
Our multi-policy reasoning engine assembles multiple candidate strategies, including ones produced by our deep-learned models, and stages an election. Each candidate is scored against the simulated futures generated by the world model, and any action that fails our safety parameters is rejected outright. Whichever strategy best handles the scenario wins control of the vehicle.
This also means every action the vehicle takes is earned and traceable to its source. That traceability scales safer deployments better than systems where decisions emerge from a single opaque model.
Reasoning Engine
Most AV systems output a single driving strategy and trust it. Ours doesn’t.
Our multi-policy reasoning engine assembles multiple candidate strategies, including ones produced by our deep-learned models, and stages an election. Each candidate is scored against the simulated futures generated by the world model, and any action that fails our safety parameters is rejected outright. Whichever strategy best handles the scenario wins control of the vehicle.
This also means every action the vehicle takes is earned and traceable to its source. That traceability scales safer deployments better than systems where decisions emerge from a single opaque model.
May Mobility’s autonomous driving system predicts pedestrian and vehicle behaviors and then reasons through a mix of deep learned policies and proven driving strategies in real time to choose the safest action.
Industry-Leading Driverless Tech
We’re building a safe, scalable driverless future by putting better AVs on the road in new markets faster and at a lower cost.
Safe
Milliseconds matter. Our system simulates hundreds of possible futures every fraction of a second and rejects any strategy that fails its safety checks.
Scalable
Smaller models, lower-cost compute and less city-specific data. Our architecture is purpose-built to expand across geographies and platforms faster and more efficiently.
Collaborative
Partnering with NTT, Toyota, Lyft, Uber, Grab, ECARX and other partners, we’re building an ecosystem of innovation — delivering Autonomy as a Service at commercial scale.
Featured Content
Featured Tech Resources
Solving the hardest problem in autonomy: handling unexpected scenarios in real-time
Conventional autonomous systems can capably handle situations they’ve trained on. But they struggle outside that training data, and the industry’s answer has been to collect more of it — at enormous cost in time, miles, and dollars.
Our approach is different. Instead of betting everything on having seen every possible scenario in advance, our vehicles reason through the scene in front of them in real time, refreshing their understanding in a fraction of a second. The result is an autonomy system that generalizes to new geographies, new road types, and complex urban environments without the billions of miles and dollars other approaches require.
Solving the hardest problem in autonomy: handling unexpected scenarios in real-time
Conventional autonomous systems can capably handle situations they’ve trained on. But they struggle outside that training data, and the industry’s answer has been to collect more of it — at enormous cost in time, miles, and dollars.
Our approach is different. Instead of betting everything on having seen every possible scenario in advance, our vehicles reason through the scene in front of them in real time, refreshing their understanding in a fraction of a second. The result is an autonomy system that generalizes to new geographies, new road types, and complex urban environments without the billions of miles and dollars other approaches require.
Learn More About May Mobility
Meet the team making a difference in the world of autonomous vehicle technology and learn more about the core values that drive us.
Learn More About May Mobility
Meet the team making a difference in the world of autonomous vehicle technology and learn more about the core values that drive us.
