Perception control for obstacle detection by a cross-country rover

Perception control consists of optimally tuning sensor or processing parameters in order to increase perception efficiency under requirement constraints or while adjusting to the environment. Perception control in the task of obstacle detection from cross-country navigation is addressed. The authors show how to maximize the vehicle safety at a given velocity, or inversely how to derive the maximum speed for a given safety. This optimization problems requires the joint analysis of how the vehicle velocity sets look-ahead requirements, how the computational cost of perception is related to the perception variables, the window of attention and image resolution, and how the reliability of the obstacle detection system is related to these variables. This criterion relies on experimental performance statistics. This system has been implemented and tested in outdoor operation.<
>