Due to the short-horizon nature of the MPPI method, it may fall into local optima, which can result in task failure. We point out that this is due to the standard MPPI method’s inability to update its mean action promptly in complex environments. Therefore, in this work, we propose utilizing a diffusion model to autonomously generate the mean action for the MPPI method, thereby enhancing its performance in complex environments. Based on the powerful representational capabilities of diffusion model, we integrate environmental point cloud data to generate the mean action, which improves the method’s robustness and adaptability in various environments. Finally, our method’s effectiveness is validated through multiple simulated experiments.