An Adaptive Preference Joint Ant Colony System for Multi-Objective Multi-agent Pickup and Delivery

Yu-Ze Li, Xiao-Fang Liu, Zhi-Hui Zhan, Jun Zhang

Keywords: Multi-agent pickup and delivery, multi-objective optimization, ant colony optimization

TL;DR: An adaptive preference joint multi-objective ant colony system (APJ-MOACS) to solve bi-objective MAPD problems, i.e., minimize the total working time and balance the workload among agents.

Abstract: The multi-agent pickup and delivery (MAPD) problem aims to allocate tasks to agents and plan the path of each agent. Existing methods tend to construct solutions by selecting agents first and then tasks or opposite. They ignore the strong coupling between task allocation and path planning, meeting challenges in terms of diversity and convergence on multi-objective problems. Thus, this paper proposes an adaptive preference joint multi-objective ant colony system (APJ-MOACS) to solve bi-objective MAPD problems, i.e., minimize the total working time and balance the workload among agents. In APJ-MOACS, two pheromone matrices are built to record the historical preference between neighboring tasks in agents for two objectives. APJ-MOACS represents solutions as a sequence of agent-task pairs and constructs a joint probability distribution to achieve coordinated optimization of task selection and agent assignment. Particularly, the selection preference of an agent-task pair is determined based on four factors, i.e., pheromone, heuristic information, matching factor, and balance factor, whose weights are adaptively controlled. The matching factor describes the matching degree between agents and tasks based on load demands. The balance factor is defined based on the attribute and current state of agents to help balance workload among agents. Experimental results on 18 instances constructed based on TSPLIB demonstrate that APJ-MOACS achieves superior convergence and diversity than state-of-the-art multi-objective ant colony optimization algorithms.

Submission Number: 56