Current powered exoskeleton (exo) control algorithms for locomotion assistance and rehabilitation are based on assistive, resistive and error augmentation paradigms. Within the assistive controller’s family, assist-as-needed consists in applying a corrective force proportional to the error (actual limb position versus reference pattern). Our final goal is to implement a fully adaptable control mechanism to allow a full lower limb exo to dynamically adapt the gait pattern to each patient. We propose to use a modified version of tacit learning algorithm in combination with a variable stiffness actuator to explore the improvement of the adaptability in comparison to stiff actuators. The preliminary results show that using this concept on a compliant actuator it is possible to modulate a fixed trajectory to adapt to the position limits that are induced by user’s movement capabilities.