Abstract: For the object grasping task in complex household scenes, a grasping method with pushing and grasping skills collaboration is proposed. Firstly, the grasping task in complex environments is formulated as a Markov decision process,with explicit definitions for variables such as state and action. Then, the pushing-grasping skill confidence evaluation network（PGCE-Net） is designed to evaluate the confidence of the grasping and pushing actions and the pushing distance in the current scene, achieving the collaboration of pushing and grasping skills. Inspired by human strategies for grasping objects in clutter environments, a clutter level evaluation network（CE-Net） is designed to infer the level of clutter in the current scene to optimize and adjust the robot actions, which effectively improves the grasping success rate. Finally, a suitable reward function is designed based on the characteristics of pushing and grasping actions, which enhances the pushing and grasping skills. The feasibility and effectiveness of the proposed method are validated through robotic grasping experiments in simulation and real environment.