Flexible team formation and task allocation for multi-robot systems in complex domains

点击翻译按钮获取中文摘要

The exploration of an unknown environment to gather information (e.g. mapping) is a problem naturally suited to the use of multiple intelligent robots. In highly challenging and dangerous environments such as urban search and rescue (USAR, where robots are used to search for human victims after a disaster) significant benefit derives from using robots with differing motor, sensory, and control abilities (heterogeneous robots). While a robot might find a crevasse it cannot enter, another team member might be small enough to move through this, for example. Similarly, different terrain is amenable to different robot types (e.g. humanoid vs wheeled). In such domains robots must also be considered expendable, meaning that team structure will not be static: new types of specialized robots may be called for, and robots that are lost or damaged must be replaced. Ideally multiple teams should be released across a broad area and collaborate to build a shared map. Existing approaches to this typically require centralized control (leaving the system vulnerable if a controller is lost), static team structures, and a shared external representation for space (e.g. GPS, which may be unavailable). Similar issues to USAR exist in space exploration, military, and security domains, as well as area coverage tasks (e.g. robotic harvesting). This work will involve the decentralized coordination of multiple teams of heterogeneous robots (wheeled, treaded, humanoid, and microrobots) in a USAR domain. It will support the gradual development of a shared representation of the world and one another's abilities through local encounters between robots (without broadcast communication). This will allow better decision making as to which tasks should be distributed to which robots (e.g. adapting to partial failure of robots, or knowing when a newly-encountered robot type might be useful). Teams will adapt their structure as exploration unfolds, adding members to acquire new skills, for example, and removing unneeded or poorly performing members (including their leaders), allowing these to be picked up by other teams (along with the knowledge they contain). All robots will function autonomously (i.e. not rely solely on a leader) to the degree their abilities allow.