Human-Machine Teaming Lab
The research goals of the Human-Machine Teaming laboratory are to develop human-robot interaction (HRI) and distributed artificial intelligence (DAI) methods that result in effective, reliable, and resilient robotic systems for real-time, mission deployments in dynamic environments.
The Human-Machine Teaming Laboratory's HRI research focuses on:
The Human-Machine Team Laboratory's DAI research focuses on:
Our research has extended existing cognitive task analysis (CTA) techniques in order to apply them to the Chemical, Biological, Radiological, Nuclear and Explosive (CBRNE) incident response and Wilderness Search and Rescue (WiSAR) domains. The Goal-Directed Task Analysis and modified Cognitive Work Analysis CTA methods were extended to accommodate these large systems in which humans are considered system components, rather than system users. The CTA elicited knowledge capturing unobservable cognitive processes, decisions, and judgments representing expert system performance that must be considered when developing a new system incorporating robots.
We have developed the Cognitive Information Flow Analysis (CIFA) to combine the results from multiple CTAs in order to understand how information will be used, communicated and transformed in a new system. Traditionally, robotic systems and HRI have been designed and developed with very little input from actual users. While researchers have begun to incorporate an analysis of users into the robot system development process, our effort is the first to apply CTA to the CBRNE domain, before beginning the robotic system design. The CTA and CIFA results are directly impacting the design and development of our system of human-robot interfaces.
The goal of our CTA efforts are to identify necessary information to guide the design and development of robotic and HRI technologies. Our work in this area can be decomposed into three primary areas.
Our HRI research goal is to develop capabilities for humans to supervise and task large, heterogeneous robot teams. The remote deployment of robots significantly complicates the presentation of meaningful, timely, and relevant information to the human operators or supervisors. Humans' ability to understand robot provided information and to react appropriately decreases disproportionately as the number and types of robots increase. Our HRI research focuses on developing a system of interfaces that provide advanced interaction and visualization capabilities to address these complexities, while supporting human decision making and situation awareness.
Our team has developed a number visualization capabilities to permit a single operator (e.g. the unmanned vehicle specialist) to effectively supervise and command large teams of robots. The majority of deployed systems rely on a human to teleoperate (specify each command to) a single robot, or small number robots. Teleoperation of large numbers of robots is infeasible. Robot autonomy is improving; however, autonomy can introduce ``out-of-the-loop syndrome'' that further complicates the human's responsibilities. Thus, new methods of tasking robots and visualizing robot provided information are required.
We are currently developing a system of interfaces that integrates the unmanned vehicle specialist interface with interfaces that support a human command hierarchy. The higher user command hierarchy levels incorporate information provided by robots and responders to support overall response decision making that is communicated to the lower command hierarchy levels and results in the allocation of tasks to heterogeneous robot teams. The CTA and CIFA results have been fundamental to designing and developing the system of interfaces.
Our multiple robot/agent coalition formation research goal is to develop theoretical foundations and a system of coalition formation algorithms to support the allocation of task coalitions composed of humans and heterogeneous robots that are robust in a range of real-time environmental circumstances. Coalition formation partitions a set of agents into different teams. The HMT Laboratory has demonstrated the limitations of applying software agent coalition formation algorithms to actual robot hardware systems and was the first to demonstrate coalition formation on real robot systems.
Coalition formation assigns robots to task teams based on specified criteria; however, choosing the optimal coalition from the set of all possible coalitions is an intractable problem. Heuristic algorithms provide good solutions, but cannot be generalized to all real-time, dynamic environments. We have proven that even approximating the coalition formation problem for task allocation is NP-Hard. The HMT Laboratory is developing a collection of algorithms that support coalition formation for a large range of real-time, dynamic situations. Task allocation is very difficult and cognitively demanding for humans, thus the coalition formation research is being integrated into the HRI research in order to support human decision making.
Our robotic situation awareness research goal is to provide robots with situation awareness that augments autonomous capabilities and improves HRI. Current research related to situation awareness focuses on humans' ability to obtain an awareness of the situation. As autonomous robotic capabilities improve, it becomes necessary for robots to understand the immediate situation in order to respond and interact properly with other robots or humans. Currently robots exhibit limited situation awareness. Robots can perceive and process information from the environment, but true comprehension of the information and projection of the information to determine future actions is often only possible for very limited domains. Developing a complete situation awareness capability requires an extensive set of system mechanisms. The HMT Laboratory has validated a preliminary robotic forgetting mechanism based on human forgetting.
The development of the forgetting algorithm is the first step towards developing a robotic situation awareness capability. We plan to to develop the capabilities necessary to obtain robotic situation awareness. Robotic situation awareness provides the potential to improve the autonomous capabilities of robots. Additionally, robotic situation awareness should facilitate mixed-initiative interaction between robots and humans. The HMT Laboratory's future endeavors will focus on how the situation awareness capabilities effect robot autonomy, and more importantly impact HRI. The hypothesis is that improved robot situation awareness will provide a broader span of control for humans, while also improving humans' understanding of robot activities.