In recent years, many works focusing on improving motion tracking accuracy of hydraulic actuators have been done by developing nonlinear control algorithms. Model nonlinearity compensation, such as valve deadband, nonlinear friction force, and flow leakages, is one of the key points in the control design. However, the inherently nonlinear valve flow characteristic is usually treated as linear for simplicity, which might lead to undesirable flow rate deviation and reduce the tracking precision especially when the cylinder moves at a low speed. In this study, an online valve flow nonlinearity compensation method is developed to improve the motion tracking precision of hydraulic actuators. Specifically, polynomials are used to mimic the nonlinear valve flow characteristic between the valve control input, pressure difference, and flow. The coefficients in polynomials are synthesized based on the offline identification experiments while the lumped discrepancy is updated online in the proposed control design. The adaptive robust control approach is also used to develop the closed-loop motion controller, which can handle the parametric uncertainties and uncertain nonlinearities in the electro-hydraulic system. Comparative experiments are conducted on an independent metering system with a differential cylinder and the results show the improved tracking precision with nonlinear valve flow comparing to the linear one, verifying the effectiveness of the proposed approach.