AI move to location with natural variances in path

I am making a game in which there is one person in VR who basically has to distinguish a few rogues (players) out of their small army of robots (AI). Anyone who has seen footage of Mass Exodus has seen how the players can blend in with the AI to a good degree because the AI bots wander around with a good amount of imperfection and randomness, including odd paths, and just stopping and standing for a few seconds.

Creating a behavior tree by which the AI just wanders to random locations and waits a random amount of time is pretty simple. My problem beings with the fact that my AI need to be commanded to go to a certain location (of a few on the map), on the fly while maintaining variation in their walking patterns instead of taking the default least distance traveled route, with human-like turning and odd behavior so it isn’t impossible for humans to try to fit in, while in the end they still go to the location and preform their task.

There are a few problems that I wont go into, like how to have the bots line up and take turns retrieving an item from a certain location, and evenly dispersing themselves around an object, but any info on tricks to make that simpler is appreciated too.

I have heard something about a neural network plugin that could provide varied behavior between individual bots when given a move to location task, but have yet to find anything like that.