Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Large-Scale Continual Scheduling and Execution for Dynamic Distributed Satellite Constellation Observation Allocation

Itai Zilberstein, Steve Chien

TL;DR

This work tackles large-scale, dynamic observation scheduling for distributed satellite constellations by formulating DCOSP as a DDCO P and introducing D-NSS, a scalable online decomposition-based solver. It establishes an omniscient offline optimal bound via a static reduction and demonstrates that D-NSS achieves near-optimal solutions with significantly reduced computation and communication compared to baselines. The approach leverages the Geometric Neighborhood Decomposition to form subproblems and employs dynamic repair to handle problem changes, enabling operation across millions of variables in real time. The results, grounded in Planet-like and Walker-like constellations and tied to NASA's FAME demonstration, indicate strong practical potential for onboard autonomous scheduling in time-sensitive Earth observation campaigns.

Abstract

The size and capabilities of Earth-observing satellite constellations are rapidly increasing. Leveraging distributed onboard control, we can enable novel time-sensitive measurements and responses. However, deploying autonomy to satellites requires efficient computation and communication. This work tackles the challenge of efficiently scheduling observations for hundreds of satellites in a dynamic, large-scale problem with millions of variables. We present the Dynamic Multi-Satellite Constellation Observation Scheduling Problem (DCOSP), a new formulation of Dynamic Distributed Constraint Optimization Problems (DDCOP) that models integrated scheduling and execution. DCOSP has a novel optimality condition for which we construct an omniscient offline algorithm for its computation. We also present the Dynamic Incremental Neighborhood Stochastic Search algorithm (D-NSS), an incomplete online decomposition-based DDCOP algorithm that repairs and solves sub-problems when problem dynamics occur. We show through simulation that D-NSS converges to near-optimal solutions and outperforms DDCOP baselines in terms of solution quality, computation time, and message volume. As part of the NASA FAME mission, DCOSP and D-NSS will be the foundation of the largest in-space demonstration of distributed multi-agent AI to date.

Large-Scale Continual Scheduling and Execution for Dynamic Distributed Satellite Constellation Observation Allocation

TL;DR

This work tackles large-scale, dynamic observation scheduling for distributed satellite constellations by formulating DCOSP as a DDCO P and introducing D-NSS, a scalable online decomposition-based solver. It establishes an omniscient offline optimal bound via a static reduction and demonstrates that D-NSS achieves near-optimal solutions with significantly reduced computation and communication compared to baselines. The approach leverages the Geometric Neighborhood Decomposition to form subproblems and employs dynamic repair to handle problem changes, enabling operation across millions of variables in real time. The results, grounded in Planet-like and Walker-like constellations and tied to NASA's FAME demonstration, indicate strong practical potential for onboard autonomous scheduling in time-sensitive Earth observation campaigns.

Abstract

The size and capabilities of Earth-observing satellite constellations are rapidly increasing. Leveraging distributed onboard control, we can enable novel time-sensitive measurements and responses. However, deploying autonomy to satellites requires efficient computation and communication. This work tackles the challenge of efficiently scheduling observations for hundreds of satellites in a dynamic, large-scale problem with millions of variables. We present the Dynamic Multi-Satellite Constellation Observation Scheduling Problem (DCOSP), a new formulation of Dynamic Distributed Constraint Optimization Problems (DDCOP) that models integrated scheduling and execution. DCOSP has a novel optimality condition for which we construct an omniscient offline algorithm for its computation. We also present the Dynamic Incremental Neighborhood Stochastic Search algorithm (D-NSS), an incomplete online decomposition-based DDCOP algorithm that repairs and solves sub-problems when problem dynamics occur. We show through simulation that D-NSS converges to near-optimal solutions and outperforms DDCOP baselines in terms of solution quality, computation time, and message volume. As part of the NASA FAME mission, DCOSP and D-NSS will be the foundation of the largest in-space demonstration of distributed multi-agent AI to date.
Paper Structure (20 sections, 10 equations, 3 figures, 2 tables, 3 algorithms)

This paper contains 20 sections, 10 equations, 3 figures, 2 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Satellite Observation Scheduling
  4. Dynamic Distributed Constraint Optimization
  5. Problem Definition
  6. Multi-Satellite Constellation Observation Scheduling
  7. Dynamic Multi-Satellite Constellation Observation Scheduling
  8. Algorithms
  9. Obtaining an Optimal Solution to DCOSP
  10. Dynamic Incremental Neighborhood Stochastic Search
  11. Baseline Algorithms
  12. Experiments
  13. Setup
  14. Satellite Constellations
  15. Downlinks
  16. ...and 5 more sections

Figures (3)

  • Figure 1: We illustrate $\Bar{h}(\delta_t)$; the line spans the entire horizon of the DCOSP $\delta$ where $\Bar{h}(\delta_t)$ shows the unknown interval the problem is static as defined by $\delta_t$.
  • Figure 2: The satellite constellations: Planet (left) and Walker (right). Dots show satellites in an orbital plane.
  • Figure 3: Results of 10 large-scale simulations for the Planet (top row) and Walker (middle row) constellation. We report the average satisfaction, total message volume, and per-agent runtime. Note the log scale for message volume and runtime. We also show the solution quality across a fixed number of iterations for iterative algorithms for instances with $v=5$ (bottom).