Abstract: For the adaptive control of nonlinear systems with any initial state error, we propose an adaptive control strategy with a modified initial state error function through learning. The algorithm divides the whole control process into many subprocesses of equal length, and it does not rectify the error in each subprocess until the system has completed tracking. Furthermore, parameter learning is performed in each subprocess to speed up the convergence. In the control process, we design a continuous controller using the arctangent function, which can avoid the chattering problem. Two examples are provided to demonstrate the tracking performance of the proposed approach through computer simulations.