<p>During the actual locomotion of a two-wheeled legged robot, its physical state parameters vary dynamically, and external disturbances can easily cause balance loss. A key challenge in balance control lies in enhancing the robot’s adaptability to variations in physical parameters and improving its robustness against disturbances. To address this challenge, by simulating physical parameter variations through changes in robot height, this paper develops an adaptive neuro-fuzzy inference system (ANFIS)-based control method to enhance the robustness of balance control in two-wheeled legged robots. The method first fuzzifies the PID controller’s gain parameters using fuzzy logic to establish a nonlinear fuzzy PID (FPID) balance controller with anti-disturbance capability under a fixed height. This controller is subsequently utilized for data acquisition. Subsequently, by adjusting FPID controller parameters across multiple heights to acquire robot state variables and output data, a robustness dataset encompassing diverse operating conditions is compiled. Finally, the dataset is leveraged to train ANFIS, resulting in an ANFIS controller capable of maintaining anti-disturbance performance across varying heights. Experimental results demonstrate that the ANFIS method exhibits superior anti-disturbance performance compared to both PID and FPID approaches when the robot operates at different heights.</p>