Aiming at the problems such as low efficiency, poor accuracy and path deviation in existing path tracking control algorithms for mobile robots, a tracking algorithm based on attention mechanism network (ACN) is designed. Firstly, a binocular vision system is deployed at the front end of the mobile robot to collect images in real time, and a transformer training model is constructed based on the self-attention mechanism, and LSTM structure is introduced into the encoder and decoder of the model to improve the model"s two-dimensional image data training ability. Weight allocation of computing resources in transformer network will determine the complexity of the model. Therefore, CS group intelligent optimization algorithm is introduced in combination with transformer to find the best weight allocation parameters in the global scope and constantly modify the learning rate through adaptive step size adjustment, so as to achieve more accurate path tracking of mobile robots. The results of 10 groups of test experiments show that the average time of the proposed algorithm to complete the path tracking task is only 9.2s, and the path deviation variance of the sampling point is 0.015, which is better than the existing robot path tracking algorithm.