Skip to Main Content
We present a "bottom up" approach, based on intelligent software agents, to real-time scheduling problems. The approach is market based; the schedule is arrived at through bidding. We recast the real-time scheduling problem as one of allocating, within stipulated deadlines, producer outputs to consumer demands over a time horizon. Producer agents submit bids to consumer agents. These bids are structured as option bids wherein consumer agents pay option premiums to producer agents thereby gaining the ability to postpone commitments. Producer agents use a game theory philosophy called "coopetition"; that is, they simultaneously compete and cooperate with other producer agents. When they compete, agents use the maximin principle from non-cooperative game theory to devise bidding strategies. When they seek to identify potential partners to coordinate bidding strategies with, agents use the Nash bargaining protocol from cooperative game theory. The framework has a strong positive feedback component in that success breeds success and only the fittest producer agents survive. Agents use memory based reasoning techniques to learn to revise their strategies as games are repeated. Agents are mobile; this enables them to conduct negotiations more efficiently by co-locating to the same machine. Agents are also endowed with limited speech recognition and speech synthesis capabilities; this facilitates interactions with the human decision maker who supervises the entire scheduling process.