|
Agent systems, such as those used to control robots, make decisions about their actions and take into account changes in the surrounding environment. The agent’s reasoning includes deliberating about its goals, such as whether to adopt an additional goal, to prioritize or reprioritize its goals, and to suspend some goals. In popular agent systems, such as those based around the Belief-Desire-Intention (BDI) architecture, deliberation is usually qualitative only, in that goals are dropped when they are found to be in conflict with other goals, or no longer believed to be possible, rather than as a means of increasing a measure of utility. In this paper we add a quantitative dimension to this reasoning process by formulating it as a Constraint Optimization Problem (COP). This allows us to incorporate preferences and other utility measures. We describe some criteria relevant to the reasoning process. The resulting model is able to encompass multiple aspects of agent deliberation, enabling the agent to make decisions that take into account more options and sources of information than it could by breaking the deliberation into components across its decision cycle.
| |
