Multiple UAV Task Allocation Using Distributed Auctions

Abstract

In this paper, we propose a distributed auction scheme for multiple miniature aerial vehicles (MAVs) performing a search and destroy mission. Although auction schemes are used commonly in multi-robot task allocation, constraints imposed by MAV characteristics creates conflicts during implementation of the allocation mechanism. The MAVs are subject to kinematic constraints and limited sensor and communication ranges. Due to these constraints, efficiently allocating tasks to multiple MAVs is a difficult problem. The distributed auction scheme presented in this paper provides a systematic procedure for task allocation to take place by resolving conflicts and considering the limitations of the MAVs. The performance of the task allocation scheme is validated using simulations for various target distributions, sensor ranges and communication ranges. The performances achieved using the distributed task allocation scheme is compared with (a) a greedy strategy and (b) using a combination of a validation process and a greedy strategy. The performance is measured in terms of the time required to accomplish the mission. The results show that the distributed task allocation scheme outperforms both strategies. The selection of targets for auction affects the performance of the task allocation. We develop both a risk-based and cooperative heuristics for target selection. The simulation results show that the performance achieved using a distributed auction with and without heuristics is similar, but the number of auction cycles required to accomplish the mission with heuristics is lower than without heuristics. Miniature unmanned aerial vehicles (MAVs) have the potential to be used as munitions in hostile environments. A futuristic battlefield may require deployment of multiple MAVs to carry out search and destroy missions. In such a scenario, MAVs search for targets and when the targets are detected an agent is assigned to attack the target. The primary requirement for these missions is that the MAVs should autonomously allocate tasks among themselves. Allocating multiple tasks to multiple agents is a difficult problem. The complexity of the task allocation problem increases because of the following constraints of MAVs: (a) limited sensor and communication ranges (b) kinematic constraints (c) MAVs continue to travel during decision-making in the search space. Because of limited sensor and communication ranges, each agent can share information only with its neighboring agents, hence the agent has incomplete information about the environment. With this incomplete knowledge of the environment the agents should generate efficient solutions. The agent must consider the kinematic constraints of the MAV during assignment. Due to constraint (c), the task allocation algorithm should have low computational overhead in generating solutions, otherwise the solution may not be effective. In order to meet these constraints the task allocation scheme should be able to make fast and efficient decisions with incomplete knowledge of the environment. The classical solution for the task allocation problem is to have a centralized system that generates the necessary commands for the MAVs. But centralized task allocation systems have limitations due to their single point failure and lack of scalability and robustness. In this paper, we present a distributed task allocation scheme based on an auction mechanism. The task allocation scheme is distributed, thus enabling the auction mechanism to be used under limited sensor and communication ranges. The actions taken by agents in a neighborhood are coordinated through the auction mechanism. The distributed auction scheme has a low computational requirement, fast decision-making capability and can be implemented with incomplete information, thus meeting the MAV task allocation requirements for implementation in dynamic environments.