interesting problem. have you tried using destination offsets?
at group start, you get the point in the middle of the group. then you calculate each actor’s offset to that point.
when you set a target point, this is the new point in the middle of the group when it arrives. so each actor gets a new destination: target point + individual offset.
stage 2 of the algorithm would be to choose a leader actor. the offsets move with the leader actor. if one of the followers has to move around a rock or anything, they speed up to reach their old position next to the leader.
edit: “stage 2” should work for turns.