Today, at the World Artificial Intelligence Conference (WAIC), NetEase's intelligent construction machinery brand, NetEase Lingdong, held an outdoor mine embodied intelligence technology launch event, officially introducing the world's first embodied intelligence model specifically designed for the loading scenarios of excavators in open-pit mines - "Lingjue". This launch aims to respond to the strategic goal of the National Energy Administration to achieve intelligent continuous operations and unmanned transportation in large open-pit coal mines by 2025, and provides a scalable technical blueprint for the "Guidelines for Intelligent Coal Mines".

"Lingjue" is the world's first "end-to-end" construction machinery embodied model. Relying on autonomous learning technology driven by multimodal data, it has completely overturned the traditional development model, achieving three technological breakthroughs:

  • Technological Path Innovation: Discarding the traditional modular development and scenario customization model, it pioneers an end-to-end integrated model, significantly improving generalization performance.

  • Data "Flywheel" Drive: Training data directly comes from real mine operation scenarios, effectively solving the problem of complex scenarios that simulation data cannot handle.

  • Comprehensive Domestication: Built on a self-developed domestic framework, core algorithms and hardware chips are fully independently controllable, ensuring technical security and supply chain stability.

In the harsh environment of the Huolinhe Open-Pit Coal Mine in Inner Mongolia, "Lingjue" has achieved a single-machine loading efficiency of 80% compared to manual operations, and about 70% of the working time requires no human intervention, successfully adapting to extreme cold, high dust, and other harsh environments as well as various types of mining trucks.

In addition, NetEase Lingdong announced two significant measures at the launch event: the first open-sourcing of the "Lingjue" dataset, and launching the **"2027 Industry Collaboration Plan"**. This plan will collaborate with main machine manufacturers such as XCMG, Sany, and Shantou Intelligent, as well as various open-pit coal mine enterprises, to promote joint R&D, scenario co-creation, and standard formulation through a technology sharing platform. The goal is to achieve unmanned operations in over 30 mines by 2027, accelerating technological inclusiveness through ecological synergy.

The end-to-end training framework of "Lingjue", called "Jixie Zixin", innovatively integrates three-stage training paradigms: video learning, expert demonstration, and reinforcement learning, endowing machines with dynamic evolution capabilities.

Currently, "Jixie Zixin" has successfully supported the application of "Lingjue" in mine scenarios and is rapidly expanding to more than 10 scenarios such as port cleaning, concrete mixing stations, and land coal sales. In the future, it will extend to broader fields such as agriculture and intelligent manufacturing.