Knowledge Management for Rapidly Extensible Collaborative Robots

It is difficult – but increasingly important – to build collaborative systems that can do a range of jobs in a range of ways in a range of conditions. Much of this difficulty stems from the broad scope of contextual and collaborative knowledge required to interact productively with humans amidst uncertainty and change. With respect to physical collaboration, human-robot interaction (HRI) studies have made progress on related problems by constraining uncertainty and dynamism, thus limiting the knowledge required. Our participation in an extreme context prevented us from adopting these approaches: The Defense Advanced Research Projects Agency (DARPA) Aircrew Labor In-Cockpit Automation System (ALIAS) program has a requirement that the prototype robotic copilot system it seeks to develop must be extensible in 30 days to function in a different, unspecified aircraft in a range of flight conditions and missions. To accommodate, we developed a knowledge management-inspired approach and system that allowed a variety of stakeholders to curate and rely on a dynamic body of flight-related knowledge. This had significant positive implications across design, test and use of the system, crucially enabling pilots to compose a robot- and human-legible plan for their interaction with the system using familiar conventions. Our contributions promise to accelerate quality development of extensible collaborative robotic systems for settings such as general medical assistance, disaster response and construction that require collaborative problem solving in highly uncertain, dynamic and variable conditions.