<p>The double lasso drive system is famous for its simplicity, dexterity and lightness, and has been used more and more in rehabilitation robots. However, the nonlinear effects such as friction torque between the rope and the lasso bring challenges to the robot control. In order to solve this problem and achieve precise control, the torque transfer model of double lasso under different load conditions is established. A test platform for transmission characteristics of double lasso was built, and the factors affecting the torque transmission efficiency of double lasso were verified. Based on torque drive model, a composite compensation controller is designed by combining RBF neural network with sliding mode control technology. Through simulation and experimental verification of rehabilitation robot, the effectiveness and tracking performance of the control system are proved, which provides a new idea for friction compensation control of multi-type rehabilitation robot.</p>