Human–Robot Collaboration: Robotic Arm Trajectory Planning with Deep Learning and DMP-based Intention Prediction

With the continuous development of Human–Robot Collaboration (HRC), an increasing demand exists for robotic arm systems that not only understand human behavioral intentions but also ensure safe and efficient coexistence with humans. This study introduces an innovative approach to robotic arm trajectory planning and optimization, integrating pattern recognition, deep learning, and computer vision technologies to address two critical issues in complex HRC tasks. First, we propose an improved Informer-based method that combines machine vision and deep learning to precisely predict human hand trajectories, providing accurate data for subsequent robotic arm trajectory planning. Second, we present an optimization strategy that combines Rapidly-exploring Random Trees (RRT*) for initial path planning with Dynamic Movement Primitives (DMP), integrating torque feedback from a robotic arm’s internal sensors to enhance trajectory smoothness and adaptability. This method not only enables automatic obstacle avoidance and optimizes trajectory smoothness but also significantly reduces a robotic arm’s power consumption by approximately 22.68% during task execution. These technological improvements facilitate more fluid and natural robotic arm movements, enhancing both efficiency and flexibility in complex HRC scenarios. Our integrated system aims to achieve precise and efficient control of robotic arms across various interaction scenarios, ensuring high sensitivity and responsiveness to human operators’ intentions while improving movement smoothness and naturalness.