5.5 Planning

A natural question to ask is whether the improvement in runtime performance comes at a high price in increased planner effort. The data show (see Figure 12) that there is a decrease in the amount of search effort the planner requires in order to instantiate local role-binding variables. There is also a dramatic reduction in the number of planning search nodes expanded during calls to the planner, both compared to the initial behavior and the baseline learning system. In other, the development of conventions is a `win-win' situation.

  

Figure 12: Planning effort decreases when actors learn conventions.