Robot Model from Seoul National University Predicts Future Scenes
Researchers at Seoul National University have developed the Dynin-Robotics model, which combines human commands and physical actions.
Researchers from Seoul National University have developed a robotic foundational model called Dynin-Robotics, which understands human instructions, predicts how a scene will look after an action, and then executes the task.
A research team led by Professor Jaeyoung Do from the AIDAS Laboratory at Seoul National University's College of Engineering designed the "Dynin-Robotics" system, which integrates language, images, video, and robot actions into a single model.
In evaluations conducted across four different real-world robot manipulation environments, Dynin-Robotics achieved an average success rate of 78.4%, outperforming benchmark models from NVIDIA and Physical Intelligence. The research has been accepted to NeurIPS 2026 and is scheduled to be presented in December.
Sources
- techxplore.com · Link