The project aims at increasing the ability to respond to large-scale disasters and manage emergencies by including robots and agents as teammates of humans in search and rescue teams. The project focuses on large teams of humans and robots that have only incomplete knowledge of the disaster situation while they accomplish the mission to rescue people and prevent fires.
The methodology to achieve cooperation within the teams will be based on the development of mental models shared by team members. The shared mental models will facilitate the interactions among robots and humans by providing a suitable level of abstraction enabling them to share beliefs, desires, and intentions as they work to accomplish their tasks.
The performance of teamwork models will be measured by comparing various task performance metrics (such as time to save people), system level metrics (such as computation time or message traffic), and amount of sharedness of the mental models. The experimental work will be conducted using the open source RoboCup Search and Rescue Simulator.
Broader impacts include integration of research results in undergraduate courses, availability of the software produced as open source, outreach activities to expose K-12 students to research issues and to excite them about using computing methods for real-world problems. The long term objective is to improve preparadeness for emergency situations, which will help saving lives and minimizing loss of properties.
The main goal of this project was to study ways of improving the ability of search and rescue teams to handle emergency situations after a large scale natural disaster. The specific aspects investigated were methods for agents/robots to form teams and the use of shared mental models by the agents/robots in a team. The main results are a set of algorithms and a quantitative assessment of the effectiveness of work done by teams, as opposed to work done by individuals, and how effective different types of mental models are in improving performance in accomplishing the tasks. By testing different ways in which teams are formed (for instance, when team composition is decided up front and does not change, as opposed to teams in which individuals can dynamically move to different teams as need arises) the effect of the team structure on performance was measured and assessed experimentally in a quantitative way. The testbed used to measure team performance is a public domain simulator for search and rescue developed by the international community for the RoboCup Search and Rescue competition. The testbed is challenging and mimics real situations where communication is limited, information is missing or uncertain, and time is precious to save lives and buildings. Since the experimental work was all done using publicly available software, the results are reproducible by others using the same software. Experimental results obtained in large scale simulations using maps of real cities and a large number of agents support the value of creating teams that have the ability to adapt to the current situation, moving agents where they are most needed. The results obtained are specially impressive in the case in which the tasks to be accomplished have a cost which increases with time, such as battling fires. Novel methods for allocation of agents to tasks of this type have been devised. The methods use estimates of how the cost of completing the tasks will grow over time and optimally allocate the agents to the places where they are most needed. This provides a principled way of allocating resources in difficult situations. Graduate and undergraduate students have worked on various aspects of the project, learning to do research, writing papers, doing demonstrations and outreach activities, and making oral presentations to different audiences. The problems studied are critical for large scale emergency disaster management, which will impact the wellbeing of society at large. As the costs for management and recovery from natural disasters skyrocket, better management will not only save lives, but will also reduce the recovery costs and decrease the inconvenience to the affected population.
