Autonomous vehicles are today a technologically achievable objective, but their complexity remains enormous. This explosion in data volume is driven by the need to expose learning systems to all possible driving scenarios.

As an illustration, Waymo reports approximately 204 million kilometers driven in real-world conditions, along with several billion kilometers in simulation.

At CES 2026, NVIDIA announced a significant breakthrough in simulation, the Alpamayo & Cosmos frameworks based on Monte Carlo–type methods and synthetic chaos, enabling a large fraction of real-world data to be replaced by simulated data. This fabulous innovation is expected to further increase the volume of data to be processed. However, driving takes place in an open world, where the number of possible scenarios is inherently vast. Hence this data growth and resulting complexity issues are not expected to be resolved any time soon. We , at Arocern (formerly Motiv AI), are addressing this challenge with a different approach, based on the modelling of prudence. We have defined a prudence metric that allows us to determine, at every instant, whether a vehicle driven by a human or operating autonomously, is behaving prudently.

Our key observation is that prudence functions (a concept we introduce) are the eigenvectors of motion control. In other words, instead of operating at the scenario level, we work independently on the elementary events that compose those scenarios.
‍
This approach results in:

• ‍A drastic reduction in the number of driving scenario examples required for learning and validation,
• The ability to perform true unit testing,
• A major reduction in hardware costs,
• And a significant acceleration of development cycles.

Arocern has developed a proprietary technology called Anticipative Prudence, using this approach. This is not an incremental optimization, but a disruptive innovation that fundamentally changes how autonomous driving systems can be trained and validated!