In the field of robotics, effectively realizing autonomous navigation for robots in unknown environments is a core research area of current robotic technology applications. The uncertainties inherent in unknown environments, such as complex terrains and dynamic obstacles, pose significant challenges to robot navigation. Traditional navigation methods have limitations when dealing with complex unknown environments and cannot meet the diverse application requirements of robots in such settings. Therefore, this paper proposes a reinforcement imitation learning-based robot navigation method with collision prediction in unknown environments, fully demonstrating the advantages of collision prediction and reinforcement learning models in such environments. This method enables robots to achieve efficient and safe path navigation. Simulations are conducted to verify the feasibility of the proposed method, further demonstrating its effectiveness and practicality. This research contributes to enriching the relevant studies on robots.