Architectural foundations for real-time performance in intelligent agents

摘要: Intelligent agents perform concurrent tasks requiring interaction with a dynamic environment, under real-time constraints. Because an agent’s opportunities to perceive, reason, and act exceed its resources, it must determine which operations when them so as achieve most important objectives. Accordingly, we view the problem of performance in intelligent control. We propose control requirements present agent architecture designed address them. Key features include: parallel perception, action, cognition processes, limitedcapacity I/O buffers best-first retrieval worst-first overflow, planning, focus attention, satisficing execution cycle. These allow trade quality for speed response goals, resource limitations, illustrate Guardian system intensive care monitoring contrast alternative architectures.