The maze problem, as a classic path planning issue, has always attracted significant attention . This paper proposes an improved bio-inspired neural network method aimed at enhancing the efficiency of multi-robot collaborative search in unknown maze environments. First, an improved GBNN model is introduced by incorporating the signal transmission characteristics of biological neurons, ensuring that robots select the correct movement path within a limited sensing range based on neuron activation values. Furthermore, a wall-following mode is introduced to help robots escape from deadlock states. Finally, simulation experiments are conducted to validate the effectiveness of the proposed method.