Rehabilitation interventions involving powered, wearable lower limb orthoses that can provide high-challenging locomotor tasks for repetitive training sessions, mainly when assist-as-needed strategies, such as adaptive impedance control, are designed. In this study, the adaptive behavior was ensured by software control of the robotic stiffness involved in the human-knee orthosis interaction in function of the gait cycle and speed. To estimate the stiffness, we analyzed the interaction torque-angle characteristics with experimental data. The speed-stiffness dependency was more evident when high stiffness values are demanded by the user's effort. Experimental evidence from five healthy subjects highlight that the adaptive control strategy provides a more comfortable, natural motion, and kinematic freedom as compared to the trajectory tracking control, allowing the user to contribute to the gait training. Future insights cover the implementation of gravitational compensation and real-time estimation and control of all inner dynamic properties of the impedance control law.