Pneumatic muscle-driven systems are widely used to facilitate the rehabilitation of patients with motor dysfunction. While iterative learning control algorithms address system nonlinearity, achieving swift and accurate control using conventional iterative learning algorithms remains challenging. This paper presents a test-based model-free adaptive iterative learning control algorithm (TBMFAILC) for gait motion tracking of the two-degree-of-freedom pneumatic ankle-foot exoskeleton. The TBMFAILC algorithm has been demonstrated to exhibit faster convergence, better convergence properties, and strong robustness against noise. The trajectory tracking performance of the TBMFAILC algorithm in practical environments is validated and compared. The results indicate that the TBMFAILC algorithm achieves an RMSE of 0.388 \(^{\circ }\) and an MAE of 0.388 \(^{\circ }\) with no load, approximately 0.365 \(^{\circ }\) and 0.367 \(^{\circ }\) with a 2kg load, and around 0.380 \(^{\circ }\) for both RMSE and MAE with a 4 kg load. The results indicate that the algorithm exhibits faster convergence and enhanced robustness. Consequently, it is suitable for enhancing the precision and stability of motion control of exoskeleton.