Toward dynamical sensor management for reactive wall-following

摘要: We propose a new paradigm for reactive wall-following by planar robot taking the form of an actively steered sensor model that augments robot's motion dynamics. postulate foveated capable delivering third-order infinitesimal (range, tangent, and curvature) data at point along wall (modeled as unknown smooth plane curve) specified angle ray from body first intersects it. develop feedback policies coupled (point or unicycle) sensorimotor system drive sensor's foveal function instantaneous data, in accord with trade-off between desired standoff progress-rate wall's curvature varies unpredictably manner unmodeled noise signal. prove any neighborhood within which accurately predicts local “shape” wall, neither will ever hit empirically demonstrate comparative physical studies active management strategy yields superior average tracking performance avoids catastrophic collisions losses relative to passive variant.