Complex Task Allocation and Execution for Teams of Multiple Autonomous Vehicles

摘要

Teams of unmanned vehicles promise to deliver significant benefits for many complex application scenarios, including ground-based reconnaissance. Such missions are most naturally described by human operators as a set of abstract requirements both for model simplicity and due to the unknown and dynamic nature of the working environment. Given the physically distributed nature of these missions and the inherent risk and uncertainty involved, autonomous teams are ideal for efficiently parallelizing the workload and providing additional robustness. We describe a distributed coordination framework for efficiently allocating a set of complex tasks among a team of robots using a specialized auction protocol. Task tree auctions enable participants to bid on tasks described at multiple levels of abstraction and, when appropriate, redecompose tasks higher up in the hierarchy. Typically, task allocation algorithms require a central mission planner to decompose the mission into primitive tasks a priori. However, such preplanning approaches can both inefficient and incapable of dealing with unknown environments and dynamic conditions. Allocations resulting from task tree auctions can provide the relevant agents with more abstract task descriptions, enabling them to dynamically and locally replan during execution. We demonstrate this ability both in simulation and on a team of autonomous outdoor robots in an area reconnaissance scenario.