I am working on a physics based fps. Astar pathfinding does not have this built in, but with a little work it can be achieved. There is a setting for the agents where you can tell Astar to never move anything. There are hooks so you can tell Astar when to process the movement and provide the results. I take these results and process my own movement by applying forces and torque. One of the challenges is dealing with things when they go off mesh. I have agents smashing into a lot of things and other objects that send them flying so it cannot be expected that they will always be on mesh. When agents go off mesh, I have them move towards the closest point on the mesh. I have a panic routine that checks how long agents are off mesh among other things. So if they get stuck or something odd then they self destruct. I use this same logic to warn the player when they are out of bounds and then destroys them if it takes too long to return. I actually use a separate “Bounds” mesh to check for on/off mesh. This mesh does not include any static/dynamic objects. So for example, an agent can get pushed onto a large rock and will be considered in bounds but otherwise the rock will be avoided in the normal path.