GridAI: дроны китайского стартапа летают под пологом леса без GNSS и больших вычислений

网格智算 (GridAI) — a Chinese startup founded in 2025 — has unveiled an intelligent platform called "GridAI大脑" (GridAI Brain) that for the first time lets drones fly autonomously beneath the forest canopy without a GNSS signal, measure trunk diameters with millimeter accuracy, and avoid obstacles on their own. The product is already being delivered in small commercial batches.

Why forests are a "no-go zone" for drones

Beneath the forest canopy, drones lose their navigation: tree crowns block the GNSS signal, and severe radio signal attenuation makes stable video transmission and real-time manual control impossible. Classical SLAM also fails here — it can't cope with a constantly changing scene texture, where branches sway gently in the wind. The industry calls such environments "fully denied" (拒止环境).

Demand, meanwhile, is enormous. According to China's National Forestry and Grassland Administration, the country's forest stock reached 209.88 billion cubic meters in 2025, with timber production at 140 million cubic meters. The UN FAO estimates the world's annual roundwood harvest at around 4 billion cubic meters — all of which requires manual stock-taking on plots and trunk-diameter measurements.

How GridAI does without big data

GridAI builds its spatial understanding on "grid-domain learning" (网格域学习) technology and trains on dozens to hundreds of examples instead of the tens or hundreds of thousands of labeled images that classical deep learning requires. This sharply reduces the demands on hardware computation and power consumption.

Instead of recognition based on two-dimensional pixels, the system "gridifies" the physical world across three levels:

• Entity grids — discrete trackable objects in the scene
• Attribute grids — properties that change over time: position, velocity, direction of movement
• Relational grids — links between objects (for example, "a nest between two branches")
• Training — 30 photos of nests on branches; on a test set of 58 images: 0 false positives, 1 miss, 98.3% accuracy
• The founder of the technology is chief scientist 代豪 (Dai Hao), with more than 30 years of experience

Because GridAI analyzes not pixels but the deep spatial structure and relationships between objects, this data doesn't depend on lighting or weather — making it more robust to the environment. Once a work zone has been mapped, a drone equipped with GridAI doesn't need any prior scanning or model-building: it builds a dynamic spatial field in real time, plans its own route, and takes measurements without an operator.

"Only an embodied AI solution that works offline directly on the

device, delivering high-precision spatial perception and real-time tracking control, can handle the task of operating in a 'fully denied' environment under the forest canopy," — Qian Min, head of marketing at GridAI.

Where the technology goes next

GridAI positions itself not as a supplier of forestry drones, but as the creator of a universal "intelligent brain." According to the company, the platform can be built into a chip module and applied in robotic arms, autonomous vehicles, and underwater robots, giving them the ability to independently perceive space, make decisions, and act.

The technology has already been tested by the Chinese Academy of Forestry (中国林科院); partners include forestry companies China Forestry Chongqing and Zhejiang Feiliu. In warehouse logistics, GridAI works with Beijing Guangjia and Googol Technology, helping robotic arms with piece-picking, packing, and parcel sorting. Separately, a strategic partnership has been signed with Shenzhen Unicom.

Why it matters

GridAI demonstrates an alternative to the "more data — more compute" mainstream: spatial perception can be built on small samples and modest hardware, running offline directly on the device. If the approach scales beyond forests, embodied AI will become more accessible for niches that heavy models can't reach because of cost and infrastructure requirements.

Frequently asked questions

What is under-canopy space, and why is it hard to fly there?

It's the space beneath tree crowns. The crowns block the GNSS signal, radio communication attenuates heavily, and swaying branches break classical SLAM — so ordinary drones can neither navigate nor transmit stable video there.

How does grid-domain learning differ from deep learning?

Deep learning needs tens to hundreds of thousands of labeled photos and relies on two-dimensional pixels. GridAI trains on 30–100 examples and analyzes spatial structure and the relationships between objects, so it's robust to changes in light and weather and requires less computation.

Where else can GridAI be applied?

The company mentions robotic arms, autonomous vehicles, underwater robots, and warehouse logistics — piece-picking, packing, and parcel sorting. The technology is presented as a universal "intelligent brain" for various hardware platforms.