Utilizing sensory uncertainty detection for robot health monitoring and performing active perception adaptation

Disturbances, uncertainties, and unconsidered environmental conditions impact the perception of autonomous off-road robots. Such conditions lead to a decrease in availability, cause a decline in safety, and reduce planning efficiency. Perception design has a significant impact on the severity of such occurrences. Active perception and systematic quality assessment allow for detecting malfunctioning sensors and sensory disturbances and responding to them. This contribution presents a behavior-based approach to recognize quality issues in data and react to them by adjusting perception and the robot’s degrees of freedom. Trials with the off-road capable truck Unimog U5023 demonstrate the approach’s performance. The robot operates within a forest under frequently changing illumination conditions and controls the exposure of the stereo cameras to increase the availability and reliability of its collision avoidance.