NVIDIA AI Day Showcases ENdotlight's Expansion in Robotic Simulation Data Business

·by Henderson·AI
NVIDIA AI Day Showcases ENdotlight's Expansion in Robotic Simulation Data Business

At the "NVIDIA AI Day Seoul 2026" held in Seoul, the topic of training robots in virtual environments received significant attention. According to a report by ZDNet Korea, the physical AI data infrastructure company ENdotlight is expanding its customized robotic simulation data business based on its technical collaboration with NVIDIA, bringing various companies' equipment and operational environments into virtual spaces, allowing robots to simulate workflows within them.

This business is called "SimReady Asset On-Demand." Here, SimReady refers to the readiness state for immediate simulation, Asset represents the 3D data of equipment and components that make up the virtual space, and On-Demand indicates customization based on customer needs. ENdotlight is currently producing and supplying flexible simulation assets such as cables to KT using this model.

The main challenge lies in objects like cables. For physical AI to enter industrial sites, merely having a robot model is insufficient; the operational targets and surrounding environments must be accurately replicated in the virtual space. Cables change shape when touched or moved, requiring consideration of not only their appearance but also their physical properties and movement patterns, making data production significantly more challenging than for rigid components.

ENdotlight to Launch Virtual Simulation Solution for Cables

ENdotlight's approach is based on computer-aided design (CAD), first restoring the shapes and structural components of the equipment required for KT's industrial environment, then supplementing the physical properties needed for cable deformation and movement to examine the behavior of robots when interacting with and moving operational targets. The production process utilizes its self-developed CAD platform TreeNixs, covering the entire workflow from model generation, assigning physical properties and joint information, generating collision shapes, to outputting simulation files.

The final product is output in OpenUSD format, allowing it to be directly built and validated in NVIDIA's robotic simulation environment, Isaac Sim. In addition to the KT case, ENdotlight also plans to provide its assets to NVIDIA SimReady Central, opening its capabilities for producing complex industrial equipment and flexible bodies to global physical AI developers. During the conference, the DexBench project, jointly participated in by ENdotlight, NVIDIA, and RealWorld, was also introduced; this is a benchmark project for assessing the fine operational capabilities of robots.

H
About the author
Henderson