Cognitively Compliant Command for Multirobot Teams

 

To accomplish the mission envisioned by Congress in 2000 in mandating that one in three combat systems be unmanned by 2015 we are working to revolutionize the way we design and interact with robots.  Autonomous behavior, particularly that of robot teams, may be difficult to understand because their algorithms may be much different from those humans would employ under the same circumstances. These differences may become critical when robots must deal with inexperienced, naïve, stressed, or fatigued humans. Current technologies fall far short of meeting the demands of combat systems.  At one extreme, single-operator control models, such as teleoperation or waypoint following have shown severe scale limitations; at the other extreme, swarm-based algorithms can manage hundreds of robots, but only for simple tasks and with very limited direct operator influence. Thus, these paradigms impose severe limitations on our ability to deploy and command the effective robot forces we need.  At the University of Pittsburgh work on this project is proceeding in three areas:

1- Control of independently acting robots- Some tasks such as responding to a target authorization request or intervening to free a trapped robot involve operating independently on a single robot.  We are exploring control architectures to allow teams of humans to respond to such requests more efficiently.

2- Control of tightly coordinating robots- Some tasks such as formation flying or simultaneous rendezvous require tight coordination because every action of one robot imposes a need to react on other robots.  This cascading dependency is typically at a time scale and complexity that does not admit human control.  Robot coordination for such tasks is typically achieved through either:

a) centralized/optimizing control- When coordination is directed toward a predefined figure of merit such as maximizing value of targets it may be difficult           to redirect the team toward other objectives or to modulate its behavior in any way.  We are seeking to develop frameworks for designing or modifying centralized algorithms to make them more amenable to following a commander’s intent.

b) biologically inspired control laws- Team coordination relying on emergent behaviors from local interactions such as swarming has many desirable characteristics particularly robustness.  It is, however, all but immune to human control.  We are exploring amorphous programming and other techniques to find ways in which human commanders may exert control over such coordination schemes.

3- Command of robots performing complex role-based tasks- While “tight coordination” is required for tasks such as formation flying more cognitively complex forms of coordination requiring plans, roles, and so forth depend on AI type programming.  We are continuing research using the Machinetta multiagent infrastructure to develop efficient methods for humans to interact with robots coordinating with such algorithms.