Flexible Scheduling for an Agile Earth-Observing Satelllite

Adrien MaillardIJCAI 2015

Introduction

Agile Earth-observing satellites represent a significant leap in orbital capabilities, possessing the ability to quickly rotate and point their instruments at specific ground targets. However, managing the schedules of these satellites is a complex optimization challenge. Traditionally, mission plans are generated entirely on the ground, where high-performance computing can solve the intricate mathematics of orbital mechanics and task prioritization.

This paper addresses a fundamental inefficiency in this traditional model: the reliance on "worst-case" assumptions. Because ground controllers cannot perfectly predict real-time variables—such as exact data compression ratios or precise energy consumption—they must plan conservatively to ensure mission safety. This approach prevents satellite failure but leaves significant operational capacity unused. The author proposes a hybrid scheduling framework that balances ground-based global optimization with onboard local autonomy to maximize resource utilization.

The Problem: The Conservative Planning Gap

In traditional ground-based planning, every task is scheduled based on its maximum possible resource consumption. For example, an image of a cloud-covered area might compress significantly more than a high-detail urban landscape, resulting in a smaller data file. However, ground planners must assume the maximum file size to avoid overfilling the satellite's storage.

While onboard computers could theoretically make better decisions using real-time data, they lack the processing power to recalculate complex attitude maneuvers or solve global optimization problems. The challenge lies in creating a system that is flexible enough to exploit real-time efficiencies while remaining predictable enough to guarantee the success of high-priority missions.

Proposed Methodology: A Hybrid Approach

The author introduces a two-pronged strategy that divides labor between ground systems and onboard software, focusing on two critical areas: data downloading and observation scheduling.

1. Data-Download Scheduling

On the ground, the system uses Squeaky Wheel Optimization and the InCELL constraint library to construct a baseline plan. This plan assumes maximum data volumes for high-priority tasks but uses "expected" (average) volumes for lower-priority ones.

To manage the resulting uncertainty, the satellite is equipped with a fast greedy repair algorithm. If a low-priority file exceeds its expected size, the satellite refers to pre-calculated "lower bounds" on start dates. This allows the onboard system to make an immediate, safe decision: if the current task threatens the start time of a high-priority download, the low-priority task is dropped. This ensures that the most critical data always reaches the ground on schedule.

2. Observation Scheduling

Calculating the physical rotations (attitude maneuvers) required to point a satellite is computationally intensive. To keep the onboard requirements low, the ground pre-calculates these maneuvers. To handle energy uncertainty, the ground performs a backward simulation from the end of the schedule to the beginning.

This simulation determines the minimum energy level required at every point in the schedule to guarantee that all future high-priority tasks can be completed. Onboard, the satellite performs a simple check: "Is my current energy above the pre-calculated safety threshold?" If the answer is yes, the satellite executes the opportunistic low-priority observation; if no, it skips the task to preserve energy for the future.

Key Insights

  • Predictability vs. Efficiency: The framework treats high-priority tasks as "guaranteed" and low-priority tasks as "opportunistic." This allows the satellite to fill schedule gaps that would otherwise be left empty due to conservative safety margins.
  • Safety via Backward Simulation: By "looking back" from the mission's end, the ground converts a complex future-prediction problem into a simple, real-time "if-then" check for the satellite.
  • Optimized Decision Sharing: The core philosophy is to leverage the ground's computing power for global optimization (considering fairness and long-term goals) and the satellite's real-time knowledge for local adaptation.

Results and Performance

The proposed hybrid scheme demonstrates several advantages over traditional and purely autonomous methods:

  • Increased Throughput: By moving away from worst-case margins, the system executes a higher number of total acquisitions.
  • Global vs. Local Optimization: The author notes that this hybrid approach can outperform pure onboard planning. While an onboard greedy algorithm might make good local choices, the ground-based component can optimize over a much larger time horizon and ensure fairness between different users.
  • Reliable Predictability: High-priority downloads remain highly predictable, meeting the strict requirements of end-users who need certain data at specific times.

Limitations and Future Directions

While effective, the current model primarily focuses on energy and data volume. The author identifies several areas for future development:

  • Thermal Constraints: Future iterations need to incorporate satellite temperature limits into the scheduling logic.
  • Maneuver Flexibility: Currently, the satellite cannot alter its physical movement onboard. The author suggests exploring the pre-calculation of "alternative attitude movements" that the satellite could choose from to further optimize energy use.
  • Generalization: The goal is to extend this "conditional execution" framework to other types of satellite missions beyond Earth observation, creating a more universal standard for satellite autonomy.

Summary updated

BibTeX
@inproceedings{maillard2015flexible,
  author = {Adrien Maillard},
  title = {{Flexible Scheduling for an Agile Earth-Observing Satelllite}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2015},
}
Figure