Collaboration with LG Electronics and OpenMind
Converting Human Action Videos into Robot Motion Data
Preemptive Learning of Hazardous Scenarios such as Falls and Collisions
Aim Intelligence released its open-source pipeline for robot learning, "Video2Robot," on December 26, 2025.
Video2Robot is the first public outcome of the "Physical AI Safety" joint research project, which Aim Intelligence is conducting in collaboration with LG Electronics and OpenMind. The three companies are working together to structurally enhance robot safety by verifying potential risks robots may encounter in real-world environments during the simulation phase.
Actual footage of collecting and converting high-difficulty motion data such as continuous backflips through Aim Intelligence's Video2Robot. Aim Intelligence
Video2Robot is an engine that converts human actions in videos into robot training data. Specifically, it uses generative artificial intelligence (AI) to safely transform scenarios that are difficult to reproduce in reality-such as close human interaction, slipping on steep slopes, or loss of balance while handling heavy objects-into robot motion data.
This approach shifts robot safety from post-incident response to the realm of preemptive learning. In real environments, it is difficult to repeatedly record data due to concerns about damaging expensive equipment and safety risks. However, with Video2Robot, it is possible to obtain data on falls and collisions without having to film these events directly, overcoming the limitations of collecting data for hazardous scenarios.
The company emphasized that it is significant for Aim Intelligence to expand its accumulated multimodal research capabilities-including text, audio, and video-from virtual spaces to address safety issues in the physical world. Aim Intelligence aims to design safety into the learning architecture from the outset, so that robots can operate safely in real-world environments.
A representative from Aim Intelligence stated, "Going forward, we will establish a 'Physical AI Safety Framework' that allows robots to proactively learn about and prepare for various unexpected situations they may encounter in real-world environments."
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
