I think what I was trying to describe does more than minimize fuel consumption — at least assuming as I sort of was trying to imply that changes in velocity are a message passing component that is part of a distributed “proof of consensus awareness” which is then employed to maximize the likelihood of involved parties converging on the same set of safe trajectories. If applied widely — I think the principle results in reducing the number of interactions where trajectories come too close for safety margins and cause involved parties to introduce extra delay due to uncertainty in the other agent’s behavior. If you modulate speeds in order to reduce the occurrence of delay-from-uncertainty, you can ensure that one party doesn’t experience any extra delay at all rather than both being likely to, which is a net win as well as win-win as the person who does stop or experience delay will wait less time in total for the obstacle to clear.