Search

Browse

• Communities
• Authors
• Year
• Titles
• Subjects

My Space

• Login

Auction-based coordination for task allocation in Drone teams

Vastardis Georgios

Full record

---

Summary

In our days, natural disasters occur on a daily basis and typically require immediate human rescue team intervention, in order to minimize casualties. In most cases, the scene conditions may be prohibitive for human rescuers to intervene and the rescue mission may be delayed, possibly with serious consequences. Such scenarios are ideal for deploying autonomous mobile robot systems, which may provide assistance to the human rescuers by autonomously searching for threatened individuals and, in general, by providing information about the current situation. In this thesis, we present the use of a multi-drone team as a support tool to search-and-rescue missions. Using drones can potentially help to reduce the search time, because they can provide useful information from various locations through their sensors. Particularly, we focus on the optimization of the task allocation problem among the drone team members by using a distributed sequential auction method. The (simulated) auction has low computation and communication requirements, is purely decentralized, and eliminates the need for a centralized auctioneer. A number of tasks (target locations) that the drones must visit for the purpose of acquiring information is communicated to all drones at the start of the process. Then, the drones localize themselves in the environment and they initiate the auction mechanism for negotiating the targets, one at a time. Within the auction, we implemented and tested different objectives for the calculation of bids, which serve a variety of team goals in different missions this project may be used on. At the end of the auction, the drones have allocated all tasks among themselves and they proceed to visiting each one of the won tasks in the already determined order. A simple, collision-free navigation method ensures that the drones will not collide during path following. The entire project has been implemented within the Robot Operating System (ROS) and is available as an open source package. All experiments were conducted on the Gazebo robot simulator in obstacle-free environments to demonstrate the results of the proposed approach on smaller and larger problems with as many as 20 drones and 50 tasks. The main advantage of the proposed scheme is the ability to run in a fully-distributed manner over a team of inter-connected drones, or other mobile robots in general, and to provide an efficient approach for task allocation, a key problem in search-and-rescue missions.

Available Files

Services