Abstarct: Physiotherapy is one of the most widely used methods for rehabilitation. Muscle force estimation has many applications in physiotherapy, rehabilitation and auxiliary devices. The force estimated by surface electromyogram (sEMG) signals is used to control the robot, diagnose the disease, determine the type of treatment and the physiotherapy method. In this study, two general classification methods including SVM, SVM-GA, Genetic Algorithm Properties Extraction, Random Forest and SVR-baxsed SVR and SVR-GA models were used to estimate muscle force by surface electromyogram signals. sEMG signals were performed during isokinetic knee contractions with a designed rehabilitation robot, electromyogram signal was recorded from quadriceps muscles, and simultaneously applied to the target force by the pressure sensor as input data and training target.baxsed on the experimental result, the SVR-GA model performeds well in estimating muscle force with 98.22% accuracy. Following is the estimated force for robot control in two methods of torque-baxsed impedance control and impedance control with voltage strategy. And by mathematical analysis the stability of the system is guaranteed. External disturbance cannot be neglected in rehabilitation robots and a controller must be used to compensate for this disturbance.