Sub-project
Research on Robotic Foundation Models
Development of core technologies for next-generation AI robots with generality and autonomy.
Robotics technology is evolving beyond performing isolated tasks toward systems that can actively respond to changing real-world environments, make autonomous decisions, and learn on their own. In line with this trend, this project aims to build a general-purpose robotic foundation model capable of integrated perception, decision-making, and control across diverse environments.
Its goal is to move beyond task-specific models and establish the core technologies for next-generation AI robots that can think, adapt, and collaborate autonomously.
In the first phase of the research, the team is focusing on securing foundational capabilities of the foundation model in three major areas: multimodal perception, general-purpose robot control, and embodied artificial intelligence. This includes developing open-vocabulary object detection and segmentation, bidirectional understanding models that integrate speech and sound information, and interface technologies that fuse data from various sensors to ensure robust perception in complex real-world settings.
At the same time, the project aims to secure adaptable and generalizable control abilities through technologies for generating and reconstructing robot actions, as well as reinforcement-learning–based decision-making models. Furthermore, the team is developing world-model–based embodied AI that allows robots to interpret situations using common sense, ethics, and domain knowledge, and to learn continuously over long time horizons.
In the second phase, the project aims to apply these core technologies to real-world environments and further advance both performance and efficiency. The team plans to implement an executable, high-level robotic foundation model by integrating various elements such as real-world sensor data processing, real-time decision-making, model quantization for energy efficiency and reduced computational resources, and precise control of robot actions.
From the perspective of embodied AI, the project seeks to establish a foundation for robots that can interact with humans naturally and learn continuously by advancing technologies in commonsense and ethical reasoning for real-world application, long-term memory processing, domain knowledge integration, and world-model learning.
This project is jointly conducted by POSTECH, Yonsei University, Korea University, and KAIST, each contributing complementary expertise in areas such as diverse robotic motion control, embodied AI, and multimodal sensor-based perception. In particular, the team is also building large-scale simulation environments and benchmark suites that connect prompt-based multimodal understanding with robot policy learning.
This research is expected to have significant impact in domains such as automated manufacturing, smart logistics, and disaster response, It is expected that this research will enable practical deployment of intelligent robots across a wide range of applications, including public safety, smart homes, and smart logistics. In particular, developing robots equipped with human-centered ethical reasoning and adaptive learning will play a crucial role in transforming robots from simple mechanical tools into collaborative co-agents.
Professor Minsu Cho of POSTECH, the lead investigator of Subproject 2, shared his aspirations
“The robotic foundation model targeted by this subproject is a core initiative for next-generation AI technologies that bring artificial intelligence—once confined to virtual environments—into the real world. At a time when AI is expected to shape the future generations of Korea, I feel a great sense of responsibility in leading a key research effort within the National AI Research Hub. At the same time, I have high expectations, as Korea’s researchers are already achieving world-class results in AI, and leading international scholars are participating in this research center. Through the work of Subproject 2, we will do our best to help Korea emerge as a central hub for robotic AI technologies.”
