 This paper proposes a novel myoelectric control scheme that supports gesture recognition and muscle force estimation. It uses a multi-task learning approach to reduce the influence of force variation on the accuracy of gesture recognition. The proposed method achieves a 27% reduction in gesture classification error and a 0.1479 RMSD force estimation accuracy. Additionally, the proposed framework is applicable in different numbers of electrodes and has a low response time delay of about 28.22 milliseconds on average. This makes it suitable for use in myoelectric prostheses and exoskeletons. This article is authored by Ruanchen Hu, Xiang Chen, Hao Yanzhang, and others.