 |
MOVERS-WORLD has several agents who are differentiated by their types: some are ``lifters'' and some are ``hand-truck operators.'' A simple rendering of MOVERS-WORLD is given in Figure 2. Agents of the same type may have differing capabilities due to attributes such as strength. There are type-specific operators, about which only agents of that type have knowledge (such as LOAD-TOGETHER for lifters and STAND-HANDTR for hand-truck operators), and there are general purpose operators that all agents can perform (such as MOVE and SIGN). The agents in MOVERS-WORLD do not engage in any communication at planning time. Rather, they plan independently, act independently, and only communicate when necessary to establish cooperation or to maintain coordination.
It is exceedingly unlikely that agents will find efficient solutions without learning in this domain, since there is no mandated cooperation or coordination strategy. The agents have common top-level goals, but they all have their own point of view on how best to proceed. Through the use of collective memory, agents develop a ``common viewpoint'' on how best to achieve a solution. That is, an agent remembers successful patterns of cooperation in which she was involved, and can use them as a basis for future interactions. In novel situations, where no such common viewpoint has been established, there is no mechanism to force the agents to do the most sensible thing; in other words, agents might refuse to assist when they should, or agree to help when they should not.