Pony.ai launches self-improving AI for autonomous driving

Pony.ai has launched PonyWorld 2.0, an upgrade to its core autonomous driving training system.

The company said the system can spot weak driving scenarios and guide human teams in data collection and R&D.

PonyWorld 2.0 is now used across its Level 4 autonomous fleet and R&D pipeline to speed model iteration as part of its reinforcement-learning-based training approach.

The platform could support other physical AI use cases, but it gave no outside evidence for that or for its claim that its AI drivers beat humans in many scenarios.

🔗 Source: Pandaily

🧠 Food for thought

Implications, context, and why it matters.

This AI upgrade could help a wider rollout

Pony.ai said it plans to grow its robotaxi fleet from about 961 vehicles to more than 3,000 by the end of 2026 1.
The rollout spans China through partners such as Chenqi Mobility, a ride-hailing joint venture backed by Pony.ai and automaker Guangzhou Automobile Group 2. Plans also cover Europe through a robotaxi launch with Uber that starts in Zagreb, Croatia 3.
Pony.ai said it wants robotaxis running in more than 20 cities by 2026. That adds work across local driving conditions and regulations 4.

Self-driving competition now hinges on scaling service

PonyWorld 2.0 ties to Pony.ai’s push for faster iteration and easier scaling. The company describes it as a self-improving world model that supports development beyond basic capability proof 5.
Pony.ai is growing through partnerships with OnTime Mobility, a Chinese mobility services company, and Uber in Europe 6, 3. The sources do not call this a capital-light strategy or say Pony.ai serves only as a technology supplier instead of a fleet operator 6, 3.
Pony.ai said PonyWorld 2.0 now runs across its Level 4 autonomous fleet and its research-and-development pipeline 5. Robotaxi work with partners such as Chenqi Mobility suggests partners can handle day-to-day service while Pony.ai focuses on self-driving software and systems 2, 5.