Pony AI has launched PonyWorld 2.0, an upgraded world model for training its autonomous driving stack. The system is designed to diagnose its own weaknesses and guide targeted data collection to support large-scale commercialisation.

PonyWorld 2.0 features a structured intention layer that allows the model to review its own decisions. The system identifies scenarios where additional learning is required and generates data-collection tasks for human teams. This process recalibrates the world model for more precise training. The system is currently being applied across Pony AI’s driverless fleet to improve safety and traffic efficiency.

Dr. Tiancheng Lou, Founder and CTO of Pony AI, said the upgrade is an important step toward a self-improving approach to autonomous driving. He stated that as AI systems become more capable, they can play a larger role in guiding their own improvement, making development more scalable over time.

Pony AI is targeting a global fleet of more than 3,000 vehicles by the end of this year. The company plans to deploy in 20 cities worldwide, with nearly half of those locations in overseas markets. The company believes that as operations grow, it is essential to improve performance consistently without regression.

Pony AI was founded in 2016 and operates across China, Europe, and the Middle East. The company develops vehicle-agnostic “Virtual Driver” technology that integrates software, hardware, and services. Pony AI believes the technical approach behind PonyWorld 2.0 could eventually apply to other physical AI training systems.