To address the challenges of real-time obstacle avoidance and precise trajectory tracking faced by underwater robots in complex submerged environments, this paper proposes an autonomous navigation method integrating an enhanced artificial potential field (APF) approach with an adaptive line-of-sight (LOS) guidance technique. Firstly, a “perception-planning-control” simulation system for underwater robots is established within the UWSim platform. Then, an enhanced APF is proposed, integrating online identification of local minima with a multi-modal virtual target escape strategy. This effectively resolves the tendency of traditional algorithms to become trapped in local optima, enabling real-time, efficient path planning for the underwater robot in a three-dimensional environment. Building upon this, an adaptive LOS backstepping controller is designed, incorporating adaptive lead distance and a power-law convergence rule. This enables adaptive adjustment of controller parameters based on navigation status, enhancing the accuracy and smoothness of three-dimensional path tracking control for the underwater vehicle. Finally, simulation validation conducted within the constructed UWSim system demonstrates that: the proposed improved APF real-time path planning method maintains stable planning performance even under dynamic-static obstacle interference scenarios, exhibiting excellent system robustness; Compared to traditional LOS, the proposed adaptive LOS inverse control method achieves superior path-tracking accuracy, reducing the mean error and the root mean square error by 48% and 14.9%, respectively.