Force controlled series elastic actuators (SEAs) arewide used components of novel physical human-robot interaction(pHRI) applications such as assistive and rehabilitation robotics.These systems are characterized by the presence of the “humanin the loop” so that control response and stability depend onuncertain human dynamics. A common approach to guaranteestability is to use a passivity-based (PB) controller. Unfortunatelyexisting PB controllers for SEAs do not define the performance ofthe torque/force loop. We propose a method to obtain predictableforce/torque dynamics basing on adaptive control and oversimplifiedhuman models. We propose a class of stable humanadaptivealgorithms and we experimentally show advantages ofthe proposed approach.