This paper presents a Human-Machine Interface (HMI) for controlling lower limb exoskeleton robot intelligently with the joint torque estimated by a new biomechanical model of skeletal muscle. Based on the Microscopic working Mechanism of Molecular Motor (4M) in sacromere, this 4M model is established to reveal the relation between the characteristics of sEMG signal and active contraction force of muscle fiber. We discuss the interaction mechanism between human body and exoskeleton via analyzing the dynamics of exoskeleton robot and knee joint respectively. The knee joint torque that generated by skeletal muscle can be predicted accurately with the 4M model via measuring the sEMG signals on muscle surface. An active strategy for controlling exoskeleton robot has been proposed to train the lower limb on human intention. Experimental results show that the operator can exercise harmoniously with the exoskeleton, which prove the validity of this theoretical biomechanical model.