Abstract: A wearable bi-directional human-machine interaction(HMI) system and its control methods are proposed to enable the user to control multi-DOF robotic hand freely and feel the gripping force from the robotic hand. A force sensory resistor(FSR) array is built to measure the forearm force myographic(FMG) signals corresponding to different hand motions of the user. A multiclass classifier is designed based on the support vector machine(SVM) algorithm to recognize the hand motions and generate motion codes to control the robotic hand movements. Moreover, sensory feedback is achieved by transforming the gripping force signals of the robotic hand into electrical stimulation signals of skin based on the principle of transcutaneous electrical nerve stimulation(TENS). Experimental results show that the motion mode recognition method based on FMG and SVM can identify 10 typical hand motions with the accuracy of above 95%. The electrical stimulation method can feed back the perception of gripping force to the body accurately and help the user to grip objects without vision.