Koopman-Accelerated Model-Based Diffusion for Real-Time Robot Control

This paper proposes bilinear Koopman model-based diffusion (BK-MBD) to address the high computational cost of conventional model-based diffusion for robot control. The method lifts robot state into a high-dimensional space only once per control step and propagates all candidates through the learned Koopman operator, enabling real-time control that was previously confined to offline settings.