Movement Primitives in Robotics: A Comprehensive Survey
This paper presents a systematic review of movement primitive approaches in robot control, with a focus on learning from human demonstrations to generate complex action sequences. Integrating chronological and systematic perspectives, it comprehensively traces the theoretical evolution of movement primitives, key technical advances—including spring-damper modeling, probabilistic coupling of multiple demonstration trajectories, and neural network applications in high-dimensional systems—and their empirical effectiveness in tasks such as grasping and throwing. The study offers an in-depth comparative analysis of prevailing frameworks, establishes for the first time a structured developmental trajectory of the field, and clearly identifies current open challenges and practical limitations, thereby providing both theoretical guidance and a practical roadmap for research in robotic motor skill learning.