Abstract: There are many elderly people with limb disorder in China，and exercise rehabilitation technology is

an effective means of rehabilitation treatment. However，the serious imbalance of doctor-patient ratio makes it

difficult for everyone to enjoy rehabilitation medical services. Impedance control technology is used for the safe

interaction between rehabilitation robot and human body. However，constant impedance control is used in the

existing rehabilitation robots，which has weak environmental adaptability and is difficult to be applied in clinical

practice. In view of the above problems，a variable impedance control method was proposed. The user's surface

electromyography signa（l sEMG）was collected as input，and his upper limb motion angle was trained and recognized by the improved random forest algorithm. With the end force sensor，the motion intention was obtained，and then the rehabilitation motion was completed by the variable impedance controller. The experimental platform was built，and the track tracking and rehabilitation training abilities were verified and evaluated. The results show that the designed controller could control the robot to complete rehabilitation movement.

      Key words: rehabilitation robot；surface electromyography signa（l sEMG）；improved random forest algorithm；variable impedance