This paper surveys the developments in the area of vision for mobile robot navigation over the past 20 years, focusing on indoor and outdoor navigation. For indoor navigation, the paper discusses structured and unstructured environments, including geometrical and topological models, optical flow, appearance-based methods, and object recognition. For outdoor navigation, it covers vision-based navigation in cluttered environments, including systems like NAVLAB, vision-guided road-following for "Autobahns," and the Prometheus system. The paper highlights significant progress in both areas, such as the ability of robots to navigate complex indoor environments and the success of NAVLAB systems in long-distance autonomous driving tests. The paper divides the discussion into indoor and outdoor navigation, each further subdivided based on the mode of vision use. For indoor navigation, three main approaches are discussed: map-based navigation, map-building-based navigation, and mapless navigation. Map-based navigation uses pre-defined models of the environment, while map-building-based navigation constructs its own models. Mapless navigation relies on object recognition or tracking. In map-based navigation, systems like FINALE use geometric models and statistical models of uncertainty to determine the robot's position. The paper describes how FINALE uses a Gaussian distribution to represent uncertainty and how it updates the robot's position using a Kalman filter. The paper also discusses incremental localization, where the robot's initial position is known and the system refines its position over time. For topological navigation, systems like NEURO-NAV use graph-based representations of the environment, with nodes representing corridors, junctions, and dead ends. The system uses neural networks to detect landmarks and follow corridors. The paper also discusses FUZZY-NAV, an extension of NEURO-NAV that incorporates fuzzy logic for navigation behavior. The paper concludes that significant progress has been made in both indoor and outdoor navigation, with advancements in model-based approaches, uncertainty handling, and the use of neural networks for navigation. The work highlights the importance of combining different techniques to achieve robust and efficient navigation in complex environments.This paper surveys the developments in the area of vision for mobile robot navigation over the past 20 years, focusing on indoor and outdoor navigation. For indoor navigation, the paper discusses structured and unstructured environments, including geometrical and topological models, optical flow, appearance-based methods, and object recognition. For outdoor navigation, it covers vision-based navigation in cluttered environments, including systems like NAVLAB, vision-guided road-following for "Autobahns," and the Prometheus system. The paper highlights significant progress in both areas, such as the ability of robots to navigate complex indoor environments and the success of NAVLAB systems in long-distance autonomous driving tests. The paper divides the discussion into indoor and outdoor navigation, each further subdivided based on the mode of vision use. For indoor navigation, three main approaches are discussed: map-based navigation, map-building-based navigation, and mapless navigation. Map-based navigation uses pre-defined models of the environment, while map-building-based navigation constructs its own models. Mapless navigation relies on object recognition or tracking. In map-based navigation, systems like FINALE use geometric models and statistical models of uncertainty to determine the robot's position. The paper describes how FINALE uses a Gaussian distribution to represent uncertainty and how it updates the robot's position using a Kalman filter. The paper also discusses incremental localization, where the robot's initial position is known and the system refines its position over time. For topological navigation, systems like NEURO-NAV use graph-based representations of the environment, with nodes representing corridors, junctions, and dead ends. The system uses neural networks to detect landmarks and follow corridors. The paper also discusses FUZZY-NAV, an extension of NEURO-NAV that incorporates fuzzy logic for navigation behavior. The paper concludes that significant progress has been made in both indoor and outdoor navigation, with advancements in model-based approaches, uncertainty handling, and the use of neural networks for navigation. The work highlights the importance of combining different techniques to achieve robust and efficient navigation in complex environments.