“The Diffusion Policy algorithm enables robotics models to learn from multi-modal demonstrations, where a task is performed in multiple different ways, while maintaining training stability.”