For each test I used 2 Roaming Targets and varied the number of roaming AI:
About 20 min. with a roaming AI to Roaming Target ratio of 6:1 yielded clustering at each Roaming Target and active Pawn-to-pawn collision avoidance when Player was within range. When a Roaming Target was finished interacting, new movement occurred toward the newly unoccupied AI for idle AI that were not actively involved in collision avoidance. During the full duration, no AI were left in an infinite idle state; it was temporary and dependent on the occupied AI’s interaction time.
About 15 min. with a roaming AI to Roaming Target ratio of 12:1 yielded similar results to the 6:1 test, except for about the 12 min. mark when roaming AI became “boxed in” due to excessive clustering around the Roaming Target. This left them immobilized and were only able to recover after a Distance Based Optimization Reset event (occurs when Player moves a specified distance from the AI).
If it is desired to have a relatively high ratio of roaming AI to AI Roaming Targets, then a possible solution could be to add some “dummy” non-AI Roaming Targets with no collision and having Hidden In Game enabled (if there is anything to render). This will allow roaming AI a location to move to when AI Roaming Targets are occupied; furthermore, non-AI Roaming Targets are not one-to-one. This is just one potential solution.
In the future I may establish a queue system for Roaming Targets or some other alternative.