In order to solve the shortcomings of single function, extremely poor use effect and high price of market rehabilitation prosthesis, a control system of the manipulator was proposed based on surface EMG signals. The system was mainly divided into two parts. The one part was a signal processing system based on Cortex-M4 series EMG signal acquisition, preprocessing, BP neural network classification. The other part was the robotic arm control system based on the Cortex-M3 series. The signal was sent by the processing system to control the robotic arm with 6 DOF to realize the display of six actions through wireless transmission. The test results show that the system can realize self-learning of six actions, the success rate is over 80%, and the system has certain application value.