Detecting and avoiding interference between goals in intelligent agents

Pro-active agents typically have multiple simultaneous goals. These may interact with each other both positively and negatively. In this paper we provide a mechanism allowing agents to detect and avoid a particular kind of negative interaction where the effects of one goal undo conditions that must be protected for successful pursuit of another goal. In order to detect such interactions we maintain summary information about the definite and potential conditional requirements and resulting effects of goals and their associated plans. We use these summaries to guard protected conditions by scheduling the execution of goals and plan steps. The algorithms and data structures developed allow agents to act rationally instead of blindly pursuing goals that will conflict.