WeRide GENESIS Unites Physical and Generative AI to Redefine Autonomous Driving Simulation

 

BACA JUGA:Tolak Pendirian Crusher, Warga Kecewa Sikap Lurah Enggan Teken Surat Aspirasi, Ada Apa?

1.  A “Lego" world with unlimited combinations

 

Leveraging generative AI, WeRide GENESIS builds highly realistic cities in minutes, accurately reproducing road infrastructure, weather conditions, environmental details, and diverse traffic behaviors from around the world.

 

Within this “Lego city”, the AI Scenarios module simulates a wide-range of key scenarios that AVs could face on real roads. This includes sudden vehicle cut-ins, unprotected left turns, emergency evasive maneuvers, pedestrian or rider intrusions, and extreme events such as fires, earthquakes, road blockages, and bad weather. The module draws on billions of kilometers of real-world driving data, and over eight years of long-tail edge cases from WeRide's public road operations, ensuring AV systems are equipped to handle complex edge scenarios.

 

2.  High-fidelity interaction with agile responses

 

High-fidelity modeling of road users is widely regarded as one of the most challenging problems for autonomous driving simulation. The key difficulty lies in moving beyond representing an "average" road user to accurately capture unpredictable behaviors in the real world – for example, a human driver abruptly cutting into the lane of an AV.

 

To address this, WeRide GENESIS introduced the AI Agents module, which builds intelligent behavior models for human drivers, pedestrians, riders, and other road users. These models realistically simulate the full spectrum of traffic behaviors, from routine driving to high-risk actions.

 

By combining intelligent scenario sampling with behavior modeling, WeRide GENESIS can simulate a wide variety of road types, traffic conditions, and the distribution of dynamic and static participants. This allows tech teams to evaluate AV decision-making safety and system robustness under different algorithm configurations, predict performance across multiple Operational Design Domains (ODDs), and continuously improve iteration efficiency and quality.

Sumber: