Throwing in the Towel : Faithless Bounty Hunters as a Task Allocation Mechanism

Bounty hunting has been shown to solve the multiagent task allocation problem robustly in noisy and dynamic scenarios with multiple other agents of unknown quality. This bounty hunting model does not require task exclusivity, but does require task commitment. We examine eliminating this second requirement, thus allowing bounty hunters to commit to tasks but abandon them later (to jump ship) for more promising tasks, without telling other agents of their intent. Bounty hunting agents must use an adaptive valuation mechanism in order to avoid too much redundancy in working on tasks. We examine how one might revise this mechanism in the face of task abandonment. We compare jumping ship favorably against other bounty hunting methods and against methods which are more “auction-like” in that they permit exclusivity, and we do so under both static environments and ones in which agents, and task scenarios, change dynamically.