Search for: All records

Abstract This paper presents a new method of detecting incipient immobilization for a wheeled mobile robot operating in deformable terrain with high spatial variability. This approach uses proprioceptive sensor data from a four-wheeled, rigid chassis rover operating in poorly bonded, compressible snow to develop canonic, dynamical system models of the robot's operation. These serve as hypotheses in a multiple model estimation algorithm used to predict the robot's mobility in real-time. This prediction method eliminates the need for choosing an empirical wheel-terrain interaction model, determining terramechanics parameter values, or for collecting large training datasets needed for machine learning classification. When tested on field data, this new method warns of decreased mobility an average of 1.8 meters and 2.9 seconds before the rover is completely immobilized. This system also proves to be a reliable predictor of immobilization when evaluated in simulated scenarios of rovers with passive suspension maneuvering in more variable terrain.