We are using a team of robots to cooperatively track and pursue enemy entities that have been detected. Unmanned air vehicles (UAVs) and unmanned combat air vehicles (UCAVs) are a growing research interest [28], led by the availability of cheaper platforms that are easier to use. The SRI UV-robotics project focuses on building a system to carry out a mission objective using a team of UGVs and UAVs. Each UGV or UAV is an autonomous agent with its own view of the world, own onboard reasoning capabilities, and own set of resources (such as power, computation, and a unique set of sensors). During a mission, there may be limited opportunity to communicate with the human controller. Therefore, the agents must rely on one another to complete the mission. Our research concentrates on providing reactive regulation of low-level sensor systems and vehicle controllers so as to attain high-level mission goals, while reacting to unforeseen circumstances and taking advantage of the evolving situation.
The UV-robotics domain resembles the SUO domain in that it requires the rapid assessment of the operational situation, the determination of the viability of existing plans and control policies, and the modification of goals and objectives based on those findings and the available resources. Unlike the SUO domain, the decisions are made by the (automated) agents themselves and the agents must negotiate solutions in a cooperative fashion. One of the challenges of UVs (or any physically mobile agent) is the need for a reactive system. Perception of, and knowledge about, events and actions in the physical world are generally imprecise. To perform tasks reliably and repeatedly requires dynamic monitoring.
Just as the SUO EA filters alerts to avoid overloading the human decision maker, we must also filter alerts to an autonomous agent to avoid overloading its computational resources. Resources are always limited, particularly on a mobile platform, so a balance must be struck between usefulness and resources used. A good example of such balance is the computational resources available onboard our robots. With an infinite number of CPU cycles, we would be able to generate large numbers of contingency plans and evaluate each with simulation. However, we have only 20% of the CPU available for robot control and monitoring. Therefore, we have to make design decisions that limit the complexity of both control and monitoring algorithms, possibly leaving extension hooks in anticipation of greater processing power in the future.