<p>Admittance control, benefiting from human-robot interaction compliance, is widely used in rehabilitation robot studies. However, using inappropriate parameters for the admittance control model can cause harm like overuse syndrome. Therefore, it is necessary to dynamically adjust these parameters to assist patients under varying recovery periods, enabling active rehabilitation training across a wider range of recovery stages. Integrating multiple intelligent approaches presents a promising solution to this challenge. This paper proposes a variable admittance control strategy that employs a variable operator fuzzy neural network (VAC-VOFNN). The VOFNN facilitates the fuzzification of the inference process, leading to superior non-linear fitting capability. Additionally, the network’s parameters are updated online to match the rehabilitation stages of different subjects. Compared to the admittance control strategy with fixed parameters (ACS-FP) and the variable admittance control strategy with fuzzy neural networks (VAC-FNN), the proposed strategy reduces the root mean square (RMS) of surface electromyography (sEMG) from the medial gastrocnemius by 29.14% and 29.04%, respectively, while also decreasing the average interaction torque by 28.63% and 12.24%, respectively. These results suggest that the proposed strategy leads to reduced effort from subjects and increased training cycles before muscle fatigue during the same rehabilitation activities. This makes it beneficial for ankle rehabilitation of patients in different recovery periods.</p>