Autonomous vehicle technology company May Mobility today announced a strategic partnership with leading mobility service platform CaoCao to explore the large-scale commercialization of robotaxi services in international markets, beginning with Europe.

Under the agreement, the two companies will conduct joint feasibility studies and commercial exploration of robotaxi deployment and look at how to advance pilot programs in key markets. The partnership will also deepen collaboration in market expansion, operational model innovation, and commercialization pathways to progress from pilot validation toward scalable deployment.

CaoCao will leverage its strengths in ride-hailing operations, fleet management, vehicle support and maintenance, and large-scale commercial operations, serving as the fleet owner and operator of robotaxi services. May Mobility’s AaaS (autonomy-as-a-service) offering will power the fleet, its in-situ autonomous driving system reasoning through unfamiliar road environments and allowing it to adapt efficiently into new international markets.

“Expanding autonomous ride-hail across new countries takes technology that scales as fast as the operation does, and that is exactly what May Mobility delivers,” said Edwin Olson, CEO and Founder of May Mobility. “CaoCao has set an ambitious goal of deploying 100,000 robotaxis by 2030, and we are excited to work together to help realize this vision.”

On June 18, CaoCao officially unveiled its RoboX strategy, marking its evolution from a mobility-service platform into a physical AI mobility technology platform for the AI era. The company believes that, as AI becomes increasingly capable of task execution, real-world mobility and fulfillment networks will become critical infrastructure connecting the digital and physical worlds. The strategic partnership with May Mobility will further advance the overseas implementation of its strategy and accelerate the development of a global robotaxi operations network.

As a global leader in autonomous driving technology, May Mobility developed a proprietary autonomy architecture that integrates deep learning, world models, and real-time reasoning to enable AVs (autonomous vehicles) to navigate complex and dynamic road environments. In partnership with Toyota, NTT, Lyft, Uber, and Grab, the company delivers AaaS at commercial scale and has completed more than half a million commercial autonomous rides across deployments in the U.S. and Japan.

 

May Mobility’s new AV architecture

Last month, May Mobility announced the launch of its fifth-generation autonomy system, which fuses deep learning, a predictive world model, and its proven reasoning engine for a radically efficient on-vehicle architecture. The company’s autonomous driving system can predict pedestrian and vehicle behaviors and then reason through a mix of deep-learning policies and proven driving strategies in real time to choose the safest action.

The company says its technology takes a fundamentally different architectural approach to modular AV stacks and pure end-to-end models. The integration of deep learning and reasoning allows the system to benefit from training data while understanding how the vehicle’s actual context may be different and reacting accordingly. The combination of these approaches enables its AVs to generalize and handle novel situations, new geographies, and complex driving conditions without the intensive data and compute requirements that can limit typical autonomy systems.

“Driving by memorization is bad,” said Olson. “Humans don’t need to see a billion miles of road to drive safely. The brain instantly builds a mental model of the world and then reasons through it. Our new system approaches driving the same way, and it dramatically changes how autonomy can safely scale.”

The latest updates enhance the company’s autonomy capabilities, which have successfully delivered more than 525,000 commercial rides and more than 1.1 million autonomous miles commercially to date, including driverless deployments in three U.S. states. On public roads, the system delivers noticeable improvements in ride smoothness and more driving confidence when navigating through complex environments.

 

World model and reasoning & planning engine

According to the company, conventional AV models trained on large volumes of driving data can capably handle situations they’ve seen before. Yet these systems may be brittle when encountering situations outside their training data. May Mobility can address such edge cases with two fused components running in tandem on a vehicle.

The first component, the company’s integrated world model, enables it to reason through complex situations outside its training data in real time. It understands the environment around the vehicle through a distillation of physics, rules of the road, and driving culture.

Applying the world model repeatedly gives the vehicle hundreds of “what if” simulations to analyze every 200 ms, each one representing a possible future. The model predicts how every road user’s behavior will affect every other, simulating up to 10 s into the future in each simulation. By evaluating the array of probable futures, the company’s autonomous technology not only matches situations to training data but also thinks through the scene and identifies the safest path.

The second component is a reasoning & planning engine. While most AV systems output a single driving strategy with no alternative to validate against, the company’s multi-policy reasoning system selects from multiple strategies, all of which compete to control the vehicle based on how well each strategy handles the simulated futures generated by the world model.

In a fraction of a second, it simulates the outcomes of deep learning and proven strategies and rejects any action that fails its safety parameters. Therefore, vehicle control is always earned and can be traced back to its source, a key distinction from end-to-end models.

May Mobility says its fifth-generation autonomy system lays the foundation for a cost-efficient system that scales effectively across autonomous ride-hail markets, challenging industry assumptions that massive datasets and custom hardware are required to achieve full autonomy. The company believes that conventional AV stacks that “memorize every situation” are expensive because of the need to collect and train on massive datasets, yielding models that can be enormous.

Its models are based on understanding how the world works, rather than memorizing everything they’ve seen, allowing them to be much smaller. That can result in lower-cost hardware that safely handles the “long tail” of complexity more effectively than may be possible using conventional approaches.

May Mobility has begun rolling out its technology update to expand the capabilities of its current fleet and enable new driverless deployments in the near future. Ride-hail networks will be among the first to experience the new technology, including May Mobility’s upcoming deployment on the Uber platform in Arlington, TX.