A Flexible Job Shop Scheduling Representation of the Autonomous In-Space Assembly Task Assignment Problem

TL;DR

This paper introduces a flexible job shop scheduling model for autonomous in-space assembly tasks, proposing both optimization and reinforcement learning solutions to improve task assignment efficiency.

Contribution

It presents a novel FJSP representation for autonomous space assembly and develops both MIP and RL formulations for task scheduling.

Findings

MIP formulation finds optimal solutions minimizing makespan.

RL approach learns interjob dynamics but does not reach optimality.

Future work aims to combine strengths of both methods.

Abstract

As in-space exploration increases, autonomous systems will play a vital role in building the necessary facilities to support exploration. To this end, an autonomous system must be able to assign tasks in a scheme that efficiently completes all of the jobs in the desired project. This research proposes a flexible job shop problem (FJSP) representation to characterize an autonomous assembly project and then proposes both a mixed integer programming (MIP) solution formulation and a reinforcement learning (RL) solution formulation. The MIP formulation encodes all of the constraints and interjob dynamics a priori and was able to solve for the optimal solution to minimize the makespan. The RL formulation did not converge to an optimal solution but did successfully learn implicitly interjob dynamics through interaction with the reward function. Future work will include developing a solution formulation that utilizes the strengths of both proposed solution methods to handle scaling in size and complexity.