Perception systems in autonomous vehicles must accurately detect and classify objects within their surrounding environments. Although challenges remain due to variations in environmental light intensity, visible cameras are widely utilized in mobile robot (MR) perception systems. Multimodal object detection (MOD) techniques that combine visible and infrared (IR) images can effectively address these issues. This paper proposes a YOLO-CSSA network, which consists of two streams to process information from RGB and IR images, building upon improvements made to YOLO-v5. The integrated CSSA blocks demonstrate exceptional two-channel fusion performance with high reliability. Furthermore, successful testing on two datasets, LLVIP and FLIR, yields superior results compared to state-of-the-art methods. However, the RGB-IR image pairs that are typically captured exhibit spatial misalignment due to sensor inconsistencies and a lack of information interchange and connectivity within the dual-branch backbone, which ultimately impairs the efficacy of RGB-IR object detection (OD). In conclusion, the proposed YOLO-CSSA network has the potential to enhance the efficiency of autonomous robot navigation in low-light environments.