Markov Decision Processes For Multi-Objective Satellite Task Planning

TL;DR

This paper introduces a semi-Markov decision process framework for multi-objective satellite task planning, enabling efficient scheduling of complex operations with multiple constraints and objectives, outperforming traditional methods in speed and reward.

Contribution

The paper presents a novel SMDP formulation for satellite task scheduling that considers multiple objectives simultaneously, improving planning efficiency and solution quality.

Findings

Comparable performance to baseline methods in single-objective scenarios

Faster solution times in single-objective planning

Higher schedule reward in multi-objective scenarios

Abstract

This paper presents a semi-Markov decision process (SMDP) formulation of the satellite task scheduling problem. This formulation can consider multiple operational objectives simultaneously and plan transitions between distinct functional modes. We consider the problem of scheduling image collections, ground contacts, sun-pointed periods for battery recharging, and data recorder management for an agile, resource-constrained Earth-observing spacecraft. By considering multiple mission objectives simultaneously, the algorithm is able to find optimized task schedule that satisfies all operational constraints in a single planning step, thus reducing the operational complexity and number of steps involved in mission planning. We present two solution approaches based on forward search and Monte Carlo Tree search. We baseline against rule-based, graph search, and mixed-integer-linear programming approaches. The SMDP formulation is evaluated in both single-objective and multi-objective scenarios. The SMDP solution is found to perform comparably with the baseline methods at greatly increased speed in single-objective scenarios and greater schedule reward in multi-objective.