Servo - August 2011

GEERHEAD

Sentries are an interesting variable in this mix of maps. If the robot knows the location of a sentry, it can determine what is or is not visible to the sentry and avoid areas where it would become visible. Because people can turn around and view in all directions, the robot assumes the sentry has omni-directional vision, or that it can see in all directions at once.

Each cell with a ray passing through it is assumed to be visible by the sentry if it is within 50 m of the sentry (the distance limitation of human sight at night). Because rays can diverge and shine light in a cell where no cost is assigned (assuming one ray alone could not light the cell), the robot calculates cells where two rays diverge by a distance of at least the diagonal width of a cell and assigns a cost there. Objects that block rays throw a monkey wrench into these types of calculations as some rays will not persist to a distance to create the divergence.

If Sensors Fail

If the robot’s sensors fail and it cannot locate the sentries, the robot will remain as covert (hidden) as possible. The robot accomplishes this using a generalized intervisibility map. This map rates the covertness of an area or cell based on its visibility from any other area or cell.

The robot’s objective is to view a target location such as a building. The robot wants to see the location without being seen. Therefore, the robot should not get any closer to the location than necessary to view it appropriately. To do this, the robot considers buildings as dangerous objects with a much higher generalized visibility cost so that it will automatically avoid getting too close to them.

The cost maps are calculated in the following order: The stealth map is calculated first, then the individual intervisibility cost map is next, followed by the generalized intervisibility cost; then the cost from light sources. The robot then chooses a course based on the cells that have desirability cost. The robot generates a path it will take and is capable of changing its path dynamically as the environment changes with objects and/or sentries appearing or changing their locations.

The robot uses a “finite state machine” to switch tasks/behaviors from goal seeking to observing to hiding to waiting. Each task or behavior determines the robot’s actions based on its current circumstances, i.e., location and environmental information. The goal-seeking behavior determines the robot’s actions while in motion. The observing behavior takes charge when a goal is reached so the robot can watch the location. The robot hides when it detects threats such as sentries. It waits when it sees that all potential paths to the goal are beyond the detection thresholds.

Figures derived from actual trials demonstrated that the robot should weigh its calculations with 80 percent for stealthiness and 10 percent for both desirability and escapability of goals or cells.

Conclusion

Future work should enable the robot to automatically weigh stealthiness, desirability, and escapability against each other based on sensor configuration, environment, and robot state. SV