This paper presents a practical localization and navigation framework for quadruped robots in outdoor pedestrian environments without requiring a pre-built metric map. The system combines GNSS-based localization with pedestrian route planning from OpenStreetMap to generate a sequence of waypoints for long-range navigation. Robot heading is estimated from successive GNSS fixes using a sliding-window PCA method with safeguards for stationary startup and low update-rate conditions. A lightweight controller tracks the route using heading and cross-track error regulation with speed ramping, while a reactive LiDAR-based finite-state module overrides commands to safely avoid obstacles and then re-joins the planned path. The approach is validated on a Unitree Go2 EDU across four heterogeneous routes demonstrating consistent goal reaching under obstacle-free traversal, isolated obstacles, and complex unmapped occlusions.