A test of my custom navigation mesh system, using Unity's physics engine to scan the environment into node data, which is then post-processed for use in A* pathfinding. This scene is a small apartment complex, (approximately 75m x 75m, with 6 floors, scan density of 0.5m) in which the full scenario calls for the player to find their way through the building to a goal on the 5th or 6th floor, then have an AI literally hunt them down by scent, while presenting the player with an aggressively realistic AI who leaves few paths for escape. The goal being to manufacture real tension.

This particular test was about determining whether I would have to prescan the entire game, or whether it was possible to generate the high detail pathing maps on the fly when encounters are required -- my goal was to also use procedural generation to produce the majority of the architecture of an empty city, and so storing pathing maps in advance would be difficult. So far, it seems to work fine, even on low-end hardware, but I need to test against better assets than my primitives. WebGL, unfortunately, doesn't support threading, so I've disabled the analysis portion.

The pathfinding supports a keyring for passage control and hazard recognition, areas of exclusion and 'nearest of N points' navigation. Various scanning tags are implemented through interfaces, so most behaviour is often as simple as adding dummy mono during scanning or referencing the interface. From there, I can write navigation data into storage, or generate on the fly in a reasonable amount of time. This is my go-to for pathfinding, as I found there were a gaps in Unity's pathfinding solution that left me a little annoyed when it came to ease-of-use.

Once the scanning is complete, clicking will order Capsule Man to path to the requested location. WSAD to move camera, handy slider to change layers.