Bounty Hunters and Multiagent Task Allocation

We propose a system for multiagent task allocation inspired by the model used by bounty hunters and bail bondsmen. A bondsman posts tasks for agents to complete, along with bounties to be collected by an agent on completion. Multiple agents, taking the role of the bounty hunters, compete to finish tasks and collect their bounties. While a task remains uncompleted, its bounty gradually rises, making it more and more desirable to pursue. Unlike auctions, this model does not assume rationality in agents’ bids (as there are none), and since tasks are not exclusive to given agents, the system is robust to highly noisy environments. We examine how agents may locally develop rational task valuations in such an environment, gradually adapting to dividing tasks according to the agents best suited to them. We compare different methods for building these valuations against approaches which are more “auction-like” in that they permit exclusivity, and we do so under both static environments...

• Type: Journal
• Year: 2015
• Where: ATAL
• Authors: Drew Wicke, David Freelan, Sean Luke
• Keywords: ATAL 2015, Intelligent Agents