I think a frame delay would be optimal, but johne5's code isn't what I am looking for, as the ally is always trying to get 'inside' the player, not stay behind him. Right now, I need to make the ally follow me at a constant delay behind me. Of course, this mechanic will also raise the problem of the ally being stuck mid-air after the player jumps over a gap, if I were to set up a basic coordinate offset, but that is something I will try to figure out later.

Once you find and touch an ally, it starts following you through the level. The player has to reach the end of the level while trying to find/collect as many 'allies' as possible. Wow, thanks for taking interest in my problem! I am trying to create a level-based 2D platformer game that uses physics. Nothing says that the object 'replaying' actions needs to be the original object that generated them. Looking into Replay systems (as in 'Action Replay', like football replays) in guides and on the UAS may help you to shortcut some of this if you need a high level of precision. The final system will depend heavily on how accurately you need to mimic these actions. For instance, 'MoveToNewLocationEvent', 'JumpEvent', 'AttackEvent'- most of these having additional information stored about the specifics of the events necessary to recreate them accurately (locations, facing directions, timings, etc). If, however, you need to repeat specific actions in sequence with a high level of accuracy, then it may be best to generate a series of character 'action events', the same way an Undo system generally works. Instead of generating the path using the NavMesh (or whatever system you're using), you instead just generate and retrieve a list of previous positions (sampled every few frames) from the 'followed' character and use that the same way. If you only need a sequence of waypoints, any guide on pathfinding is going to be directly applicable.

I'm with methos5k, the specific use-case here is really important.

It follows 2 slowly