Hesai Technology Partners with Ropedia to Develop Embodied Intelligence Data Products

2026-11-16 14:59
Favorite

en.Wedoany.com Reported - On November 16, Kosmo, the 3D world data engine under China's Hesai Technology, entered into a strategic partnership with Ropedia, a physical AI infrastructure company. The two parties will jointly carry out large-scale production of high-quality real-world data, covering spatial data acquisition, structured processing, standardized delivery, and model validation. The Ropedia Xperience 2 research project, jointly supported by both parties, was released on November 12, producing a dataset comprising 1,589 3D scanned scenes, 74 manipulation tasks, and 4,230 first-person recordings, with a total recording duration of 131 hours.

The partnership clearly defines the technical division of labor between the two parties. Kosmo is responsible for spatial data acquisition and 3D reconstruction of real environments, while Ropedia uses HOMIE Gen2 and the Experience Engine to complete motion recording, 4D reconstruction, and semantic annotation, associating spatial geometric information with human manipulation behavior data to form multimodal data for embodied intelligence model training and evaluation.

All 1,589 3D scenes in the Ropedia Xperience 2 project were scanned using Kosmo One and uniformly registered with dynamic manipulation recordings. The research team has validated four data reuse pathways, including converting real scenes into interactive simulation environments, retargeting human motions into robot trajectories, generating robot behavior data through virtual teleoperation, and automatically generating long-horizon task trajectories.

The research project also conducted model evaluations on depth estimation, 3D object detection, and spatial intelligence tasks. The research team integrated real-environment modeling and human behavior recording into a unified data system, using the acquired scene and manipulation data to support robot policy training, simulation testing, and model performance evaluation.

The two parties plan to provide multimodal spatial data products to global large model, embodied mobile manipulation model, and world model R&D teams, and to adjust subsequent acquisition and delivery processes based on issues identified during model training and validation.

This bulletin is compiled and reposted from information of global Internet and strategic partners, aiming to provide communication for readers. If there is any infringement or other issues, please inform us in time. We will make modifications or deletions accordingly. Unauthorized reproduction of this article is strictly prohibited. Email: news@wedoany.com