Movement Primitive

Movement primitives are compact mathematical representations of robot skills, learned from demonstrations and used to generate and control robot movements. Current research focuses on improving the generalization and adaptability of these primitives, exploring model architectures like Dynamic Movement Primitives (DMPs), Conditional Neural Movement Primitives (CNMPs), and diffusion-based methods, often incorporating neural networks and manifold learning techniques to handle complex motions and diverse contexts. This work aims to create more robust and versatile robotic systems capable of performing intricate tasks in unpredictable environments, with applications ranging from surgery and manufacturing to human-robot interaction.