Localization involves using sensor data (like odometry, IMU, and lidar scans) and comparing it against a pre-built map of the environment. Algorithms like AMCL (Adaptive Monte Carlo Localization) are commonly used to estimate the robot's pose (x, y, theta) with a certain degree of uncertainty. Accurate localization is the bedrock upon which all other navigation tasks are built.

Before a robot can navigate autonomously, it needs a map of its environment. This map can be generated using sensors like lidar or depth cameras. ROS provides tools like gmapping and cartographer for SLAM (Simultaneous Localization and Mapping), allowing robots to build maps while simultaneously tracking their own position within them.

The navigation stack employs two main types of path planners: global and local. The global planner (e.g., Dijkstra, A*) generates a long-term, collision-free path across the entire map. The local planner (e.g., DWA, TEB) then takes this global path and generates short-term, dynamically feasible velocity commands, reacting to immediate obstacles and ensuring smooth motion.

While path planning accounts for static obstacles in the map, obstacle avoidance deals with dynamic elements or unforeseen barriers. The local planner plays a critical role here, using sensor data to adjust the robot's velocity and trajectory to prevent collisions. This often involves techniques like velocity obstacle methods or dynamic window approaches.