Skip to main content

WeRide's New AI Model Learns to 'Understand' the Real World

At the 2026 World Artificial Intelligence Conference (WAIC), WeRide dropped a new model called WIIT. It's not just another AI that processes data—it's designed to actually understand the real world. The core idea? Break down the messy, ever-changing environment into tiny, checkable "facts." WeRide calls these "Minimum Physical Fact Units."

So, what does that mean in plain English? Imagine a self-driving car approaching a crosswalk. A traditional AI might just detect a pedestrian. WIIT, on the other hand, would extract facts like: "A person is standing at the curb," "The traffic light is red," and "The car is moving at 30 mph." Then it reasons about how these facts relate, verifies them against reality, and orchestrates a safe decision—like slowing down.

WIIT comes with four core skills: fact extraction (picking out key elements from chaos), fact reasoning (connecting the dots), fact verification (checking if those dots are accurate), and fact orchestration (using the facts to plan actions). It's like giving AI a detective's toolkit: observe, deduce, double-check, then act.

WeRide says the real challenge for physical AI isn't just processing information—it's grasping spatial relationships, behavioral logic, and dynamic changes. For example, a robot in a warehouse needs to know not just that a box is there, but that it's fragile, that another robot is approaching, and that the shelf might wobble. WIIT aims to handle that kind of complexity.

This isn't just about autonomous driving. WeRide sees WIIT powering all sorts of embodied AI—robots, drones, maybe even smart home systems. By building a cognitive framework based on physical facts, they're hoping to make AI that doesn't just see the world, but truly gets it.

Key Points

  • WIIT is a physical AI foundation model that breaks the real world into "Minimum Physical Fact Units."
  • It has four core capabilities: fact extraction, reasoning, verification, and orchestration.
  • The goal is to help AI understand spatial relationships, behavior, and dynamic changes in real-world scenarios.
  • Applications include autonomous driving, robotics, and other embodied AI systems.