LG Electronics and a consortium of industry and academic partners, including Maum AI and KT, have begun developing core models for “Physical AI” tailored to industrial environments such as manufacturing facilities.
On June 9, South Korea's Ministry of Science and ICT and the Institute of Information & Communications Technology Planning & Evaluation (IITP) held a project kickoff meeting for the Physical AI Leading Technology Development Program at LG Sciencepark in Seoul's Magok district. The initiative aims to localize key Physical AI technologies and lay the foundation for South Korea to become a global leader in the field.
The project was awarded to a consortium led by LG Electronics following a competitive bidding process. The consortium includes Maum AI, Holiday Robotics, Robotis, CrowdWorks, Alchera, KT, KAIST, Seoul National University, and the Telecommunications Technology Association (TTA), bringing together 10 organizations from industry, academia, and research sectors.
The Ministry and the LG-led consortium aim to achieve world-class Physical AI performance within two years by developing a World Model and a Robot Foundation Model (RFM), which together serve as the “brain” of Physical AI systems.
A total of 49.7 billion won ($36 million) will be invested through next year, including 34 billion won in government funding and 15.7 billion won contributed by consortium members. The goal is to create Physical AI capable of perceiving, reasoning about, and autonomously acting within the physical world. The first version of the World Model is scheduled for completion by August.
The consortium seeks to maximize real-world simulation capabilities through the World Model and improve transfer learning performance using the RFM. The ultimate objective is to boost real-robot task success rates by more than 20 percentage points compared with systems that do not utilize a World Model, surpassing the current global benchmark of 14.5 percentage points achieved by OpenGVLab.
To accelerate development, the consortium will establish a validation pipeline covering World Model training, RFM integration, testing and performance evaluation, case analysis, and retraining. The technology will undergo continuous verification and refinement, culminating in pilot deployments at real manufacturing and logistics sites to generate commercially viable outcomes.
Each participating organization has clearly defined responsibilities in the consortium. LG Electronics will lead development of the World Model, while KT will take charge of the Robot Foundation Model (RFM).
LG Electronics plans to leverage its LG AI Research Institute's AI model “EXAONE” to develop and connect the World Model and RFM. It will also handle simulation-based and real-world validation, as well as testing advanced technologies such as the World Action Model (WAM).
Maum AI will work on World Model development, commercialization testing in logistics operations, and parallel development of next-generation foundation models including WAM. KT will focus on linking and enhancing the World Model into the RFM developed by the consortium.
Robotis will develop proprietary robotic hardware and generate high-quality behavioral datasets. Holiday Robotics will build a domestically developed simulator, including its own physics engine. CrowdWorks and Alchera will establish an integrated Physical AI data collection platform and gather core datasets.
KAIST and Seoul National University will develop and validate optimal World Model architectures using three parallel methodologies: physical law consistency modeling, rare-case augmentation, and lightweight predictive modeling. The Telecommunications Technology Association (TTA) will support data standardization and Physical AI model evaluation.
Kim Young-joon, head of LG Electronics' AI Research Center, said that global tech companies' general-purpose World Models such as Cosmos and Genie have limitations in reflecting the complex working conditions and processes of domestic manufacturing environments. He added that LG aims to support immediate industrial innovation by developing an advanced World Model that improves through learning from diverse real-world errors.
Ryu Jae-myung, Second Vice Minister of Science and ICT, emphasized that securing core Physical AI infrastructure independently is the starting point for becoming a global leader in the field. He added that the government will fully mobilize its policy resources to provide strong support for the initiative.
