In the era of convergent robotics and artificial intelligence, robust visual perception is crucial for autonomous decision-making and effective human-machine collaboration. However, end-to-end learning-based methods often struggle with dynamic scene deblurring due to the absence of edge-guided features and efficient fusion mechanisms, leading to residual blurring artifacts. To overcome these limitations, we propose the Densely Edge Gating Attention Network (DEGANet), a novel architecture tailored for dynamic scene deblurring. Specifically, we design an Edge Guidance Gating Residual Block (EGGRB) to extract and refine high-frequency edge structures, thereby enhancing spatial detail restoration. In addition, a Dual-Path Multi-Scale Attention Block (DMAB) is introduced to strengthen contextual representation and improve robustness to diverse blur patterns. To further enable effective information propagation, we develop an Across-Scale Attention Fusion Block (AAFB) that facilitates adaptive feature exchange between encoder and decoder stages. Extensive experiments conducted on the widely used GoPro and HIDE benchmarks demonstrate that DEGANet achieves superior performance compared to state-of-the-art methods, both quantitatively in terms of PSNR and SSIM and qualitatively through clearer textures and sharper edges. The proposed framework generalizes well to real-world scenarios, making it particularly suitable for applications in autonomous systems, robotic vision, and intelligent perception under challenging conditions.