In this paper, we propose an alternatively adaptive finite-time (FT) parameter estimation method for a developed reaction force-sensing series elastic actuator (RFSEA). Specifically, a compact RFSEA is first designed for exoskeletons, where the spring and drive components are configured in parallel. In contrast to classic estimation approaches depending on the predictor errors to derive the adaptive law, the proposed method is developed directly based on the parameter estimation errors, which is extracted by applying a low-pass filter operation. Moreover, a feasible way to achieve the PE condition verification online is provided. Furthermore, the homogeneous correction term is used for designing the adaptive law such that the FT convergence can be ensured. Finally, the effectiveness validation of the proposed method is provided by comparative simulations for the developed RFSEA.