To accelerate the Robotic Innovation Research Scheme, our Hub conducts frontier research at the intersection of Artificial Intelligence, Robotics and Autonomous Systems. Our mission is to develop the next generation of embodied intelligence by advancing world models, autonomous reasoning and collaborative physical AI. We focus on enabling intelligent agents to understand and predict the dynamics of the physical world, perform generalized tasks in complex environments and collaborate effectively with humans and other agents. Through this research, we aim to establish the foundational technologies that will drive the future of Physical AI and real-world intelligent systems.

We developed a bisimulation-based formal planning framework that enables a humanoid robot to safely perform diverse manipulation tasks with no retraining. Using a single symbolic model built once from the robot dynamics, the Unitree G1 successfully performs object grasping, elevator-button pressing, and berry harvesting across different environments. Only the task specifications and environment labels need to be updated, enabling rapid adaptation while providing formally certified collision-free motion plans.