Abstract: We present a method for autonomous lidar-based navigation of an unmanned surface vehicle (USV) through plane fitting in an underground drainage box culvert, where the satellite and ground communication signals are limited, and lighting conditions are poor. The scenario model of the underground drainage box culvert is built based on the lidar imaging principle. To address the problem of the lidar point cloud in the box culvert being disturbed by the reflection of water waves, an improved eigenvalue-based plane fitting method is proposed, which effectively filters out the point cloud noise and obtains appropriate plane parameters. The virtual horizontal plane is used to calculate the relative positions of the USV and wall on both sides of the box culvert using the fitted plane parameters, and the desired route is planned. Based on the scenario model, simulation is conducted by using the LOS guidance law and a PD heading controller. The results show that the proposed method achieves high fitting accuracy at a low computational cost and that the determined navigation parameters are accurate and stable, allowing the USV to navigate autonomously along the center line of the box culvert. Experiments are used to validate the effectiveness and feasibility of the proposed method.