Aiming at achieving bright and clearly focused images under low-light conditions with illumination attenuation in autonomous driving scenarios, this paper proposes the MLMamba model for fusing visible and infrared images. Unlike most existing methods that extract features using convolution and directly concatenate them into a single-path network, a process that limits the preservation of complementary information and often neglects the effects of illumination attenuation. The proposed method utilizes a multi-branch Mamba structure, which is enhanced by an attention mechanism to extract both shallow and deep features from individual modality inputs via multi-scale convolution and Mamba blocks. Subsequently, a novel Mamba-based Multi-branch Fusion Module (MABM) effectively captures long-range dependencies and integrates complementary features, which can yield a richer fused representation. Finally, the improved lightweight module is used to complete the reconstruction. Considering weak light attenuation, a novel intensity loss term is devised to enhance image quality under low‐light conditions. Extensive experiments demonstrate that MLMamba achieves performance that is comparable to or even superior to existing methods on various evaluation parameters. These results validate its potential for robust, real-time autonomous driving applications.