This paper presents a model-based scheme for controlling an industrial robot without using any external sensors for physical human-robot interaction. The scheme consists of two major modules: force observation and compliance control. First, an accurate dynamic model involving the friction and rotor inertia of the robot is constructed and identified. Then, a novel generalized momentum combining the second-order moments and rotor inertias is developed. This generalized momentum contributes to estimating the external force without information on the joint acceleration, where the external force is transformed into the joint space under Jacobian operators. Finally, independent position-based impedance control is adopted to implement the physical interaction. Compared with the traditional force estimation methods, the presented scheme only requires joint angles and joint torques, which makes it more accurate in force estimation, more stable and universal for commercial robots. Simulations and experiments on the ROKAE XB4 platform validate the effectiveness of the proposed scheme.