An artificial muscle is well known to have strong asymmetric hysteresis characteristics, which depend on the load applied to the muscle. It is therefore difficult to achieve a high control performance and robustness to various loads. In a previous study, model-free adaptive control (MFAC), which is a data-driven control method, was applied to muscles, and a high tracking control performance was achieved. However, MFAC requires numerous design parameters that are extremely time-consuming to tune. To solve these problems, this study considers the tuning of the design parameters of the MFAC by introducing virtual reference feedback tuning, which is a data-driven control method. In addition, control experiments with five loads were conducted to verify the robustness of the load. The experimental results show that the proposed method achieves a high tracking control performance without an overshoot and is highly robust to loads while reducing the time-consuming routine for parameter tuning and modelling.