Quantitative evaluation of coarse-to-fine loading strategies for material rehandling

Autonomous handling of piled materials is an emerging topic in automation science and engineering. A central question for material rehandling tasks (transporting materials that have been assembled in piles) is “where to dig, in order to optimise performance”? In particular, we are interested in the application of autonomous wheel loaders to handle piles of gravel. Still, the methodology proposed in this paper relates to granular materials in other applications too. Although initial work on suggesting strategies for where to dig has been done by a few other groups, there has been a lack of structured evaluation of the usefulness of the proposed strategies. In an attempt to further the field, we present a quantitative evaluation of loading strategies; both coarse ones, aiming to maintain a good pile shape over long-term operation; and refined ones, aiming to detect the locally best attack pose for acquiring a good fill grade in the bucket. Using real-world data from a semi-automated test platform, we present an assessment of how previously proposed pile shape measures can be mapped to the amount of material in the bucket after loading. We also present experimental data for long-term strategies, using simulations based on real-world 3D scan data from a production site.