This is the latest version of this item.

Abstract

Soil compaction, an increase in soil density and decrease in porosity, has a negative effect on crop yields, and damaging environmental impacts. Mapping soil compaction at a high resolution is an important step in enabling precision agriculture practices to address these issues. Autonomous ground-based robotic approaches using proximal sensing have been proposed as alternatives to time-consuming and costly manual soil sampling. Soil compaction has high spatial variance, which can be challenging to capture in a limited time window. A multi-robot system can parallelise the sampling process and reduce the overall sampling time. Multi-robot soil sampling is critically underexplored in literature, and requires selection of methods to efficiently coordinate the sampling. This paper presents a simulation of multi-agent spatial sampling, extending the Mesa agent-based simulation framework, with general applicability, but demonstrated here as a testbed for different methodologies of multi-robot soil compaction mapping. To reduce the necessary number of samples for accurate mapping, while maximising information gained per sample, a dynamic sampling strategy, informed by kriging variance from kriging interpolation of sampled soil compaction values, has been implemented. This is enhanced by task clustering and insertion heuristics for task queuing. Results from the evaluation trials show the suitability of sequential single item auctions in this highly dynamic environment, and high interpolation accuracy resulting from our dynamic sampling, with avenues for improvements in this bespoke sampling methodology in future work.

Available Versions of this Item

• Agent-Based Simulation of Multi-robot Soil Compaction Mapping. (deposited 27 Oct 2022 10:32)
  • Agent-Based Simulation of Multi-robot Soil Compaction Mapping. (deposited 23 May 2023 12:50) [Currently Displayed]