Generative and Greedy Routing Heuristics for Heterogeneous Multi-Robot Harvesting under Fuel and Size Constraints

Document Type

Article

Publication Date

1-1-2026

Abstract

Coordinating heterogeneous multi-robot systems for agricultural harvesting presents significant challenges due to operational constraints, dynamic environments, and varying robot capabilities. This paper addresses these challenges by proposing two practical solution strategies: (1) a constraint-aware greedy heuristic that iteratively balances workload while respecting robot capabilities and operational limits, and (2) a structured Large Language Model (LLM)-based framework for generating adaptive routing heuristics. In the considered scenario, heterogeneous robots traverse narrow crop rows while satisfying size and fuel constraints, with the objective of minimizing overall task completion time. The greedy heuristic generates an initial task allocation and iteratively optimizes routes for improved efficiency, whereas the LLM-based framework formalizes prompt refinement as a reproducible process, enabling systematic generation and evaluation of routing heuristics across different LLMs. Experimental results show that the proposed greedy heuristic achieves solution quality comparable to that of the MILP formulation solved with Gurobi for small- and medium-scale instances and reduces the makespan by up to 19% compared with the layered heuristic on large-scale instances. The LLM-based heuristics provide competitive solution quality with substantially lower computation time. These results highlight the potential of both methods for real-world applications, as they can be adapted to various problem scenarios and provide valuable insights for addressing constrained coordination challenges in heterogeneous multi-robot systems beyond agricultural automation.

Publication Title

IEEE Access

Link to Full Text

Share

COinS