Research on safety strategy for human-robot collaboration based on CNN-BiLSTM-attention motion prediction

The rapid advancement of intelligent manufacturing has enabled collaborative robots not only to perform tasks autonomously but also to collaborate synergistically with human operators across diverse applications. This study presents a human-robot collaborative safety strategy based on arm motion prediction to ensure worker safety and improve task efficiency in collaborative environments. A Kinect-based arm motion acquisition method incorporating bone-length constraints is developed. An arm motion prediction approach based on a Particle Swarm Optimization (PSO)-enhanced Long Short-Term Memory (LSTM) neural network is constructed. Subsequently, a human-robot collision detection method is implemented based on the predicted arm motion trajectories. Furthermore, an experimental system is established to achieve low-latency prediction of hand motion target points, thereby supporting safe human-robot collaboration. The stability of the proposed safety strategy is validated through arm motion prediction, and an obstacle avoidance method is proposed to ensure safe robotic operations while adapting to environmental variability in assembly tasks.