This paper presents a fully integrated autonomous navigation system for wheeled-legged robots, combining adaptive locomotion control, mobility-aware local navigation planning, and large-scale path planning. The system uses model-free reinforcement learning (RL) and privileged learning to develop a versatile locomotion controller that enables efficient and robust locomotion over various rough terrains. The controller is tightly integrated with a learned navigation controller through a hierarchical RL framework, enabling effective navigation through challenging terrain and various obstacles at high speed. The system was validated through autonomous, kilometer-scale navigation missions in Zurich, Switzerland, and Seville, Spain, demonstrating the system's robustness and adaptability. The system addresses several challenges in autonomous navigation, including hybrid locomotion, smooth and efficient navigation, and seamless integration of locomotion and navigation modules. Hybrid locomotion remains a challenging area in legged robotics, requiring effective gait selection for different terrains. The system's locomotion controller, trained using model-free RL and privileged learning, enables adaptive gait selection and efficient terrain negotiation. The navigation controller, trained using hierarchical RL, enables effective navigation through complex environments. The system's navigation controller is mobility-aware, considering the robot's dynamic characteristics for efficient navigation. It processes multiple input modalities, including the hidden state of the locomotion policy, terrain height values, and a sequence of previously visited positions. The controller is trained in a simulation environment that dynamically generates new navigation paths for each episode, optimizing the learning process. The system's performance was evaluated in various urban environments, demonstrating its ability to navigate complex terrains and obstacles. The robot achieved an average speed of 1.68 m/s with a mechanical COT of 0.16, significantly outperforming traditional legged robots. The system's ability to navigate around pedestrians, avoid obstacles, and adapt to different terrains highlights its robustness and adaptability. The system's hierarchical controller enables the robot to dynamically adapt its gait based on the terrain, demonstrating versatility in navigating complex paths. The controller's ability to handle various surfaces, including grass, sand, or gravel, is attributed to the privileged training of the locomotion controller. The system's performance was compared to a conventional navigation approach, demonstrating its superiority in terms of failure rate, collision rate, planning time, and tracking error. The system's ability to explore new areas and dynamically adjust to changing situations highlights its effectiveness in complex environments. The system's integration of mobility-aware navigation planning and hybrid locomotion contributes to its ability to navigate challenging terrain and obstacles while ensuring efficient and fast navigation. The system's robustness, adaptability, and efficiency hold great promise for transforming last-mile delivery and addressing the challenges of urban mobility.This paper presents a fully integrated autonomous navigation system for wheeled-legged robots, combining adaptive locomotion control, mobility-aware local navigation planning, and large-scale path planning. The system uses model-free reinforcement learning (RL) and privileged learning to develop a versatile locomotion controller that enables efficient and robust locomotion over various rough terrains. The controller is tightly integrated with a learned navigation controller through a hierarchical RL framework, enabling effective navigation through challenging terrain and various obstacles at high speed. The system was validated through autonomous, kilometer-scale navigation missions in Zurich, Switzerland, and Seville, Spain, demonstrating the system's robustness and adaptability. The system addresses several challenges in autonomous navigation, including hybrid locomotion, smooth and efficient navigation, and seamless integration of locomotion and navigation modules. Hybrid locomotion remains a challenging area in legged robotics, requiring effective gait selection for different terrains. The system's locomotion controller, trained using model-free RL and privileged learning, enables adaptive gait selection and efficient terrain negotiation. The navigation controller, trained using hierarchical RL, enables effective navigation through complex environments. The system's navigation controller is mobility-aware, considering the robot's dynamic characteristics for efficient navigation. It processes multiple input modalities, including the hidden state of the locomotion policy, terrain height values, and a sequence of previously visited positions. The controller is trained in a simulation environment that dynamically generates new navigation paths for each episode, optimizing the learning process. The system's performance was evaluated in various urban environments, demonstrating its ability to navigate complex terrains and obstacles. The robot achieved an average speed of 1.68 m/s with a mechanical COT of 0.16, significantly outperforming traditional legged robots. The system's ability to navigate around pedestrians, avoid obstacles, and adapt to different terrains highlights its robustness and adaptability. The system's hierarchical controller enables the robot to dynamically adapt its gait based on the terrain, demonstrating versatility in navigating complex paths. The controller's ability to handle various surfaces, including grass, sand, or gravel, is attributed to the privileged training of the locomotion controller. The system's performance was compared to a conventional navigation approach, demonstrating its superiority in terms of failure rate, collision rate, planning time, and tracking error. The system's ability to explore new areas and dynamically adjust to changing situations highlights its effectiveness in complex environments. The system's integration of mobility-aware navigation planning and hybrid locomotion contributes to its ability to navigate challenging terrain and obstacles while ensuring efficient and fast navigation. The system's robustness, adaptability, and efficiency hold great promise for transforming last-mile delivery and addressing the challenges of urban mobility.