面向人机协作装配的动作识别与序列预测方法OA

In human-robot collaborative assembly,the ability of robots to respond flexibly and accurately to human actions is critical to ensuring assembly quality.To address the challenges posed by non-standard actions and dynamic changes in assembly sequences,this paper proposes a collaborative method integrating unknown action recognition and operation sequence prediction.First,for untrained actions that may occur in actual assembly,an open-set recognition strategy based on a random forest classifier is designed,incorporating a posterior probability threshold to reject unknown actions.In a speed reducer assembly experiment,the method achieved recognition accuracy of 95%for known actions and 98.65%for unknown actions.Second,to cope with the diversity and real-time variation in assembly sequences,an improved hidden Markov model(HMM)combining incremental learning and a forgetting factor is developed,enabling rapid adaptation and efficient prediction of new assembly patterns.With a forgetting factor of 0.7,the prediction accuracy reached 97.22%.Experimental results demonstrate that the proposed method maintains high recognition and prediction performance under complex collaborative scenarios with action uncertainty and sequence variability.In conclusion,the method improves the safety and efficiency of collaboration and is highly generalizable to other dynamic and variable human-robot collaborative assembly tasks.