In continuous motion, based on surface electromyography ( sEMG) signals, exoskeleton robots and humans collaborate in motion control. Muscle fatigue will affect the flexibility and robustness of human-machine collaborative control. This article innovatively proposes the use of Entropy of Incremental Fuzzy Entropy and constructs a fatigue characterization model, and objectively divides the stages of muscle fatigue; Collect sEMG signals of twelve muscles in the lower limbs during repeated continuous leg lifting movements, propose a method based on the variability sensitivity coefficient SVR to determine muscle fatigue sensitivity, achieve effective muscle selection for this movement, reduce data dimensions, propose an adaptive threshold action segmentation method based on mean squared product, segment the complete signal and extract a single action signal sequence, and analyze and calculate the fatigue trend through this model. The experimental results of the subjects show that the model proposed in this paper has a more obvious gradient feature for muscle fatigue characterization compared to time-domain and frequency-domain algorithms, and has better fatigue characterization ability compared to fApEn and FFDispEn. Davies Bouldin Index for fatigue level clustering is 0. 39. This provides a reference for improving the collaborative control of exoskeletons and achieving phased compensation assistance for fatigue.