A fundamental requirement of non-player characters (NPCs) in today's computer games is to be able to move through a complex virtual world in an intelligent way. Pathfinding or path planning techniques are used by games developers to determine suitable routes, during gameplay, from a starting location to a goal position. These techniques make use of graphs to efficiently represent the game world. Navigation graphs can be created using different constructs, such as waypoints, navigation meshes and grids. Typically, path planning systems use the navigation graph to find the shortest path between locations and do not adapt in-game based on a player's experience of navigating a path. In previous works we have explored approaches to in-game action prediction and tactic adaptation through machine learning inspired approaches. In this paper we present an approach to the in-game adaptation of NPC movement using navigation mesh cells. The technique allows NPCs to adapt their routes through cells based on previous experience whilst also preserving smooth paths. The system is applied to FPS game scenarios using the Unity3D game engine. The results show that the technique offers a promising approach to the adaptation of NPC movement through navigation meshes.