Multi-agent task allocation for harvest management

Abstract

Read online

Multi-agent task allocation methods seek to distribute a set of tasks fairly amongst a set of agents. In real-world settings, such as soft fruit farms, human labourers undertake harvesting tasks. The harvesting workforce is typically organised by farm manager(s) who assign workers to the fields that are ready to be harvested and team leaders who manage the workers in the fields. Creating these assignments is a dynamic and complex problem, as the skill of the workforce and the yield (quantity of ripe fruit picked) are variable and not entirely predictable. The work presented here posits that multi-agent task allocation methods can assist farm managers and team leaders to manage the harvesting workforce effectively and efficiently. There are three key challenges faced when adapting multi-agent approaches to this problem: (i) staff time (and thus cost) should be minimised; (ii) tasks must be distributed fairly to keep staff motivated; and (iii) the approach must be able to handle incremental (incomplete) data as the season progresses. An adapted variation of Round Robin (RR) is proposed for the problem of assigning workers to fields, and market-based task allocation mechanisms are applied to the challenge of assigning tasks to workers within the fields. To evaluate the approach introduced here, experiments are performed based on data that was supplied by a large commercial soft fruit farm for the past two harvesting seasons. The results demonstrate that our approach produces appropriate worker-to-field allocations. Moreover, simulated experiments demonstrate that there is a “sweet spot” with respect to the ratio between two types of in-field workers.

Keywords

• task allocation
• automated harvest management
• multi-agent system
• agent-based simulation
• applied AI