Covert Robotics: Covert Path Planning in Unknown Environments

1 Introduction Covert Robotics is an interesting research problem which we have presented earlier in [Marzouqi and Jarvis, 2003; Marzouqi and Jarvis, 2004]. The aim of the research is to create autonomous mobile robots with the ability to accomplish different types of navigation missions covertly, i.e. without being observed by one or more, possibly hostile, sentries operating within the same environment. A covert path is defined here as a path with the minimum exposure to being observed by sentries anywhere in the free-space of the environment. There are many applications where covert robotics systems can be useful, such as in police and military fields for surveillance, spying, and crime prevention missions. Covert missions are usually dangerous and can put human life at risk. Robots can be used to minimize such risk. Moreover, robust covert robots may perform better than humans can do, especially when real time and complex decision making is required. Finding an optimal path between two points has been given a lot of attention in the robotics field, where the goal is to minimize distance, time or power. On the other hand, there has been very little work on planning covert paths although demand is increasing for robots to participate in high risk situations where the covert behavior is required. In [Teng et al., 1993; Ravela et al., 1994], a research has been conducted for covert navigation in natural terrain environments, where a robot vehicle makes use of the terrain elevation to maintain its covertness given a known sentry location. Another work has been done in [Birgesson et al., 2003] where a robot navigates in an unknown environment to a specified destination while taking advantage of the discovered obstacles to stay hidden from a stationary known observer. We have presented in [Marzouqi and Jarvis, 2003] a solution that allows a mobile robot to navigate covertly from one point to another in a known, static environment where sentries' locations are unknown and they might be moving around. Obstacles are used as the only means to stay unobservable, they being assumed opaque and higher than all agents' sight levels. The basic idea was to assign, prior to navigation, a visibility cost to each free-space location. As sentries' locations are unknown, it is assumed there could be a sentry standing at each free-space location. Given that, the visibility cost at a point is the number of sentries visible to it (i.e. …