May Day Vision (06651): Formally reached a strategic cooperation in the aerospace field with Huantian Intelligence

Zhitong Finance App News, May Day Vision (06651) issued an announcement. Recently, the company and Huantian Smart Technology Co., Ltd. (Huantian Smart) formally reached a strategic cooperation in the aerospace field. Huantian Intelligence is a leading domestic commercial space enterprise focusing on the overall development and in-orbit operation of space-based Earth observation and remote sensing satellites. It is also the largest commercial space entity in the southwest region with the ability to build and operate a fully autonomous and controllable remote sensing satellite constellation. The two companies will cooperate on the collection of high-precision satellite remote sensing data and the reconstruction of “physical-grade usable” data assets, and jointly customize the world's first commercial remote sensing satellite, the Earth Clone Star “ECS-1,” designed specifically for physical AI training and simulation applications. This cooperation marks the first in-depth joint construction between China's physical AI industry and commercial aerospace in the satellite payload layer.

Earth Clone Star “ECS-1” is the world's first commercial remote sensing satellite specially customized for physical AI data requirements, and has many industry-leading indicators. The ECS-1 satellite is equipped with a 0.5 meter ground sampling distance (GSD) optical imaging payload with a ground-to-ground imaging width of ≥60 km; the satellite uses an advanced dual-line array stereoscopic sweep imaging system and has stereoscopic mapping capabilities. A single single-track transit observation can generate high-precision digital surface models (DSM) and digital elevation models (DEM). Comparatively speaking, with traditional commercial optical remote sensing satellites, 3D terrain reconstruction with similar accuracy usually requires joint calculation of multi-track transit images to complete.

Comprehensive efficiency: (1) Significant optimization of 3D terrain reconstruction efficiency: Relying on the monorail stereoscopic simultaneous imaging configuration, the satellite can complete regional stereoscopic mapping in a single transit. Compared with the traditional single-line array optical satellite multi-track splicing modeling model, the regional 3D reconstruction operation cycle is significantly compressed; (2) data processing efficiency multiplied: combining the company's 51WORLD MODEL world model with 4DGS (four-dimensional Gauss splash) AI processing pipeline, the target is the end-to-end cycle from satellite shooting to deliverable scenarios; the target is the industry's traditional sky level to hour (3) data acquisition cost; Significantly Reduction: The comprehensive acquisition cost of real 3D data on a city to provincial scale can be reduced by about 90% compared to traditional aerial oblique photography.

As far as project progress is concerned, ECS-1 has recently completed the review of the construction plan and officially initiated the bidding process. It is planned to complete the satellite launch networking by the end of 2027.

For a long time, it has been difficult to fully unlock the commercial value of commercial satellite data, mainly due to certain technical difficulties between original remote sensing images and the training and simulation requirements of physical AI. As a leading enterprise in the field of physical AI simulation in China, the company has full stack capabilities to quickly transform raw geographic remote sensing data into “physical-grade usable” simulation scenarios. In this collaboration, satellite-side data collection and physical AI scene reconstruction will form a direct pipeline, evolving from “sensing the Earth” to “understanding the Earth.”

Based on the self-developed physical AI engine and the 51WORLD MODEL world model, the company undertakes the core functions of “understanding” and “generation”: through multi-modal perception and large-scale spatial semantic learning, the model can unify ECS-1 and existing multi-source heterogeneous data from satellites, aviation, and ground to achieve cross-scale reconstruction from pixel-level images to the semantic digital world. Relying on the 4DGS (four-dimensional Gaussian splash) neural rendering pipeline, the original stereoscopic images sequentially completed stereoscopic matching and DSM/DEM generation, sparse point cloud reconstruction and densification, 4DGS neural rendering, full-scene semantic segmentation and element-level labeling, standardized semantic system mapping, physical attribute binding (such as friction, reflectivity, passability, carrying capacity, etc.), and finally produced “understandable, interactive, and trainable” “physical-level assets” right out of the box. It can be directly called by downstream physical AI applications.

In addition, the two sides plan the SAR satellite ECS-2, equipped with a fully polarized synthetic aperture radar and InSAR deformation monitoring capability. The imaging width is 40 km, and it relies on microwave cloud fog penetration characteristics to achieve all-weather observation throughout the day; when ECS-1 optical imaging fails due to cloud cover, ECS-2 co-monitors to ensure uninterrupted data collection in key regions.

ECS-1 satellite data can provide stable underlying data support for the physical AI business of the company's entire product line, expand commercial aerospace application scenarios, empower fields such as digital twins, autonomous driving simulation, robot training, low-altitude economy, geological emergency response, natural resources and major engineering management, and make up for existing capacity shortcomings and obtain cost advantages. In the future, it can support the company to deploy space computing power, and build a complete closed loop of integrated intelligent sensing and application of heaven and earth by extending the terrestrial computing power system to near-Earth orbit.