WeRide Unveils New AI Model for Autonomous Driving
Autonomous driving technology firm WeRide has officially launched its self-developed "physical AI cognitive foundation model," named WeRide WITT. This model aims to enhance AI's understanding of the real world by decomposing complex scenarios into identifiable "physical facts."
Autonomous driving technology company WeRide has launched its proprietary "physical AI cognitive foundation model," called WeRide WITT. This new model leverages visual and linguistic large model capabilities to deepen AI's comprehension of complex, real-time driving scenarios.
WITT, an acronym derived from "World Intelligence Toward Truth," addresses the challenges in autonomous driving by simultaneously processing video, images, and text. The model breaks down continuously changing real-world situations into discernible and verifiable "physical facts." This structured approach establishes a new generation of AI understanding frameworks centered on physical facts.
The model features four core capabilities: fact extraction, reasoning, verification, and orchestration. These functions enable detailed recognition of traffic behaviors, analysis of event causality, and ultimately, the collection of reliable data for continuous learning. WeRide states that WITT significantly reduces error rates compared to general large language models for autonomous driving applications, offering up to 98% cost savings on token expenses and 200 times higher processing efficiency than traditional methods.