Auction and Swarm Multi-Robot Task Allocation Algorithms in Real Time Scenarios

Abstract

A group of several autonomous robots (multi-robot system) can perform tasks that with only one of them would be impossible to carry out or would take much more time, moreover, they are more robust and even can be cheaper, etc than systems with a single robot. In general, the problems that have to be solved to benefit from all these advantages are divided into three main stages: task division/planning, task allocation and motion planning. Task division stage, consists on dividing the general and complex mission into simple tasks that can be carried out by a robot and, if it’s necessary, scheduling those simpler goals. The task allocation step will select the best robot or group of robots to execute each goal. Finally, the motion planning issues involve the robots’ motions coordination to get the assigned tasks. Although these steps have been explained as independent and sequential stages, they are tightly connected and the decisions made in one level affect the whole system performance. For example, in Zlot & Stentz (2006) the authors proposed a task allocation method that combined planning, task decomposition and task allocation. Although this and similar efforts can be found in the literature. Each one of these steps is still an open problem. This paper will be focused on multi robot task allocation (MRTA) issues, without taking into account the other problems. Task allocation is one of the main problems in multi-robot systems, very especially when the tasks must be executed before deadlines, that is, in real-time scenarios. In most cases this problem is an NP-hard problem, and therefore nowadays there is not any algorithm that in a reasonable computing time gives the optimal tasks allocation. Two main paradigms have been proposed in recent years to try to manage this problem in both real-time and non real-time scenarios: swarm and auction methods. At present, there does not exist any study that compares both strategies when the robots must carry out tasks with soft or hard real-time restrictions. Other works, for example Kalra & Martinoli (2006), compare auction and swarm methods but using tasks without deadlines. Therefore, the first objective of this work is to compare all those methods under different scenarios to identify the weak points of each paradigm when a deadline is assigned to the tasks. Firstly, our work presents and study three auction-like strategies based on existing ones: Sequential Unordered Auction (SUA), Earliest Deadline First Auction (EDFA) and Sequential Best Pair Auction (SBPA). In all these cases there is a central auctioneer who receives the bids from all the robots and decides the allocation of the tasks. These strategies differ between them on the way the auctioneer announces the tasks to the robots and on the