Uploaded July 2026 | Updated November 2026, 5 days ago
🦾 Introducing WITT (World Intelligence Toward Truth) – WeRide's Physical AI Cognitive Foundation Model.
Powered by Atomic Physical Facts (APF), WeRide WITT builds trusted understanding of the physical world as it transforms operational experience from vast volumes of real-world data into structured knowledge by integrating 4 core capabilities:
1️⃣ Fact Extraction: Identifies Atomic Physical Facts from real-world driving data, including traffic behaviors, multi-agent interactions, and complex physical conditions.
2️⃣ Fact Reasoning: Analyzes events, behavioral relationships, and evolving risks while enabling natural-language search and retrieval of long-tail scenarios.
3️⃣ Fact Verification: Evaluates outputs across six dimensions to reduce hallucinations and validate conclusions against observable reality. Today, WITT achieves an average factual error rate approximately one-third that of leading general-purpose AI models in autonomous driving scenario understanding tasks.
4️⃣ Fact Curation: Routes data into the most effective learning workflows, ensuring every piece of operational experience contributes to model improvement.
📉 WITT lowers token costs by up to 98%, processes up to 10,000 minutes of vehicle-operation video per day on a single GPU, up to 200× greater data-processing efficiency, and generates 100+ dynamic tags from a single request
Together with WeRide GENESIS, our proprietary general-purpose simulation model, WITT forms a Physical AI flywheel that continuously converts real-world experience into model intelligence, accelerating the development of both autonomous driving and broader Physical AI applications! 🌎🚘
#WeRide #WITT #PhysicalAI #AI #AutonomousDriving #EmbodiedAI #Robotaxi #Innovation #AutonomousVehicles #FutureOfAI
🦾 Introducing WITT (World Intelligence Toward Truth) – WeRide's Physical AI Cognitive Foundation Model.
Powered by Atomic Physical Facts (APF), WeRide WITT builds trusted understanding of the physical world as it transforms operational experience from vast volumes of real-world data into structured knowledge by integrating 4 core capabilities:
1️⃣ Fact Extraction: Identifies Atomic Physical Facts from real-world driving data, including traffic behaviors, multi-agent interactions, and complex physical conditions.
2️⃣ Fact Reasoning: Analyzes events, behavioral relationships, and evolving risks while enabling natural-language search and retrieval of long-tail scenarios.
3️⃣ Fact Verification: Evaluates outputs across six dimensions to reduce hallucinations and validate conclusions against observable reality. Today, WITT achieves an average factual error rate approximately one-third that of leading general-purpose AI models in autonomous driving scenario understanding tasks.
4️⃣ Fact Curation: Routes data into the most effective learning workflows, ensuring every piece of operational experience contributes to model improvement.
📉 WITT lowers token costs by up to 98%, processes up to 10,000 minutes of vehicle-operation video per day on a single GPU, up to 200× greater data-processing efficiency, and generates 100+ dynamic tags from a single request
Together with WeRide GENESIS, our proprietary general-purpose simulation model, WITT forms a Physical AI flywheel that continuously converts real-world experience into model intelligence, accelerating the development of both autonomous driving and broader Physical AI applications! 🌎🚘
#WeRide #WITT #PhysicalAI #AI #AutonomousDriving #EmbodiedAI #Robotaxi #Innovation #AutonomousVehicles #FutureOfAI










