Download or read book Robust, Real Time, and Scalable Multi-agent Task Allocation written by Elad Kivelevitch and published by . This book was released on 2012 with total page 215 pages. Available in PDF, EPUB and Kindle. Book excerpt: Assigning a group of mobile agents to perform certain tasks defined by their locations and priorities is a major research problem in optimization in general and particularly in multi-robot systems. This problem can be formulated as variants of the well-known Multiple-Depots, Multiple Traveling Salesmen Problem (MDMTSP), in which the agents start their travel from an initial location (a depot), and the goal is to obtain a solution that either minimizes the total of all tour lengths (the classical case), minimizes the length of the longest tour (the min-max case), or maximizes a profit of performing tasks taking into account the cost of travel to these tasks (the profitable case). It should be noted that while these problems are variants of the same problem, they are usually solved by specialized algorithms tailored for each specific variant. In this work, agents interact in a virtual economic market that can be used to solve all three variants of the problem with minor changes to the algorithm. The motivation behind using a market-based approach is mostly because of the ability of economic markets to get good results, in a short span of time, for large-scale problems, and in complex and dynamic situations. The market-based solution developed in this work demonstrates near-optimal results for all three variants, compared to state-of-the-art algorithms, each tailored for a specific variant. Moreover, the ability to obtain near-optimal results in the face of changes is demonstrated in scenarios that include the appearance and disappearance of tasks and the addition and removal of agents. The approach is further enhanced by introducing hierarchies into the market, where tasks are geographically clustered to expedite the search in large problems. Finally, the market-based approach is also enhanced using fuzzy logic to reduce the sensitivity of the cost of travel to tasks, whose location is uncertain. This work makes several new contributions to the field. First, while market-based solutions to the MTSP have been proposed in the past, this approach is the first one to include hierarchical markets and presents better and faster results with fewer constraints than previous approaches. Second, this is the first algorithm that is both as near-optimal as algorithms tailored for each specific MTSP variant and that can handle all three variants of the MDMTSP like other meta-heuristic algorithms. In particular, this solution obtains results that are very close, in terms of optimality, to the results obtained by linear programming for the classical case and the MTSP with constant profits case, but at a significantly reduced calculation time. For the min-max case, the current solution improves over the best known solution in terms of cost, by as much as 75\%, albeit at a longer runtime. Finally, this solution is scalable to large problem instances and can be used as an online algorithm that handles changes in the scenario during runtime. As a whole, this novel approach is everything required from an algorithm for assigning mobile agents to tasks: near-optimal, versatile, robust to scenario changes, and fast.