[Dynamic] Have all agents contribute to a global consensus on rewards, with a small alpha

[dmfrodrigues]

Agents with the same goal already help each other at determining the reward from one point to the goal, with a very high value of alpha = 0.5.

But all vehicles use all roads... So maybe it is possible for all vehicles to arrive at some sort of consensus on the individual rewards of each action for all agents, since that reward does not depend on the individual goals of each individual.

This would likely make learning faster, but would not improve result quality.