A new article, “The Current State of Real2Sim2Real: From On-Site Scanning to Robot Validation Environments”, has been published on Kudan Tech Blog.
Real2Sim2Real is a development approach in which a real-world site is 3D-scanned to build a simulation environment, robots are validated in that environment, and the results are then brought back to the site. The article surveys technology trends in this field and presents Kudan’s initiatives and current progress. It covers the full process: converting sites into 3D data with handheld scanners, automatically building validation environments using generative AI, generating large numbers of environments with varied conditions such as lighting and object placement, and validating robots in simulation.
With the spread of Physical AI, demand for on-site data is growing rapidly. To meet this demand, Kudan will provide a technology foundation that leverages the spatial perception technology and customer base it has built through both its robotics and digital twin businesses. The initiatives introduced in the article fall within three areas set out by this foundation: improving the efficiency of data acquisition, enabling data augmentation, and ensuring data quality. In its mobile robot development, Kudan confirmed that, for issues encountered on a real robot, the process from identifying the cause to confirming the countermeasure can be carried out in simulation with almost no manual intervention, significantly improving development efficiency. Going forward, Kudan aims to connect the insights gained from its own development to technology services, tools, and implementation support for customers, and to expand its business in the Physical AI domain.
Article: “The Current State of Real2Sim2Real: From On-Site Scanning to Robot Validation Environments”
https://zenn.dev/kudan_jp/articles/a0cb75384f4a83?locale=en
Kudan Tech Blog (Japanese)
https://zenn.dev/p/kudan_jp
