Adaptive environment observation for distributed agents moving in a lattice world

In a distributed cooperative system, each agent has to observe its environment in order to create an adequate subgoal. The paper discusses tradeoff between cost of observation and achievement of the shared final goal of agents moving in a lattice world. The model is referred to as the restricted Tower of Babel. A cooperation mechanism called LM-DMax is evaluated through simulation against observation interval, the number of agents and moving area site. We find that the model has an optimal observation interval, an optimal area size to minimize the total cost. Then three adaptive policies which modify the observation interval of each agent are proposed. They are geometrical policy, arithmetical policy and slow-start policy. Slow-start policy shows the best result among them. Though the discussion is restricted to the specific model, it is directly applicable to a vehicle scheduling problem and a robot planning problem