Visual Multi Feature Fusion for Precise Robot Grasping

Traditional robotic grasping methods that rely on a single visual feature often fail to adapt to complex environments involving variable illumination, target occlusion, or textured backgrounds. These limitations significantly reduce the robustness of grasping performance. To address this, the present study proposes a robotic grasping approach based on visual multi-feature fusion, integrating color, shape, and texture information to enhance target recognition and decision-making reliability. A weighted fusion model dynamically adjusts the importance of each feature based on scene conditions. Deep learning techniques are employed to optimize both feature extraction and fusion strategies, thereby improving adaptability and accuracy. Experimental results demonstrate that the proposed method exhibits strong robustness and generalization across a variety of challenging scenarios, including uneven lighting, occlusion, and dynamic environments.