Xpeng (NYSE: XPEV) presented a new autonomous driving framework called X-Mind at the CVPR 2026 Workshop on Foundation Model Deployment for Embodied Intelligence, according to a company statement.

Xianming Liu, Head of Xpeng Group's General Intelligence Center, outlined the company's World Model roadmap, with X-Mind described as a Predictive World Model framework designed to allow vehicles to simulate future scenarios before executing driving decisions.

X-Mind uses a Visual Chain-of-Thought process to enable autonomous systems to anticipate traffic changes through internal simulation, rather than reacting solely to current conditions. The framework includes three core components: Thought Sketch, which creates a cognitive representation combining Bird's-Eye-View layouts and driving data while reducing computational load; Recurrent Block Diffusion, which generates future scene simulations within a single forward pass to address latency issues; and Visual CoT visualization, which shows how the model predicts obstacle movements and traffic conditions before producing driving decisions.

The company says X-Mind was trained on hundreds of millions of real-world driving data frames and is designed to operate with low inference latency on automotive-grade chips.

X-Mind follows earlier releases in Xpeng's World Model series, including X-World, X-Foresight, and X-Cache. The company states that X-Mind completes its Physical AI foundational model roadmap alongside those prior frameworks.

Xpeng, founded in 2014 and headquartered in Guangzhou, China, develops electric vehicles and in-house driver assistance and powertrain systems. The company operates offices in Beijing, Shanghai, Silicon Valley, and Amsterdam.