It is difficult to accurately estimate the time delay between the transmitted signal and echo signal in the process of obstacle avoidance of the transformer robot fish. A time-delay estimation method based on the improved variable step size LMS (Least Mean Square) algorithm is proposed. Different from the traditional BCC (basic cross-correlation) algorithm, the LMS algorithm does not need prior statistical knowledge of signal and noise and has better applicability. The algorithm adaptively adjusts the output and weight vector of the filter according to the least mean square error criterion and the fastest descent method. In the iteration process, the variable step size is used instead of the traditional fixed step size to improve the convergence speed. Finally, the Sinc function interpolation is performed on the weight vector to obtain the delay estimation result. Simulation results show that the proposed algorithm has better anti-noise performance and more obvious peak time delay estimation than the traditional BCC algorithm.