Risk Prediction of a Behavior-based Adhesion Control Network for Online Safety Analysis of Wall-climbing Robots

Risk analysis in combination with terrain classification is a common approach in mobile robotics to adapt robot control to surface conditions. But for climbing robots it is hard to specify, how the robotic system and especially the adhesion is affected by different surfaces and environmental features. This paper will introduce the climbing robot CROMSCI using negative pressure adhesion via multiple chambers, adaptive inflatable sealings and an omnidirectional drive system. It presents the used behavior-based control network which allows the balancing of adhesion force, but fails in extreme situations. Therefore, a risk prediction has been developed which evaluates behavioral meta-data and allows an estimation of current hazards caused by the environment. This prediction is used to perform evasive actions to prevent the robot from falling down.