Adaptive expert systems for indirect coverage control

Herds of livestock, when left to their own devices, will forage for food in predictable patterns, and will often overgraze preferred areas while leaving other areas untouched. The field of grazing management looks to improve the efficiency of land use by moving the animals through different pastures at regular intervals, akin to coverage algorithms used in robotics but with non-uniform coverage and additional constraints due to the animals´ natural behaviors. The knowledge of the field of grazing management is largely in the hands of ranchers and other domain experts, and as such has not been the subject of much computational study. In this work, we have created novel learning techniques for expert systems that use both off-line and online learning to generate efficient performance. The rules of the expert systems are initially developed using an evolutionary algorithm, and after deployment, an adaptive algorithm tracks the state of the system and updates rule weights to improve both coverage efficiency and animal stress levels. We present a series of results based on increasingly complex versions of simulated herd models that show improvements over unconstrained motion in each case, and suggest how the algorithm can apply to robotic coverage problems.