1 of 17

�Dynamic Mission Planning Performances �for Agile Earth Observing Satellites �with Adaptive Multi-Agent System

IWPSS 2025

30/04/2025

Benjamin Marchand^1,2, B. Francesconi^1,2, A. Girard^1,2, E. Kaddoum³, A. Perles³

¹ IRT Saint-Exupéry, Toulouse, France,

² Thales Alenia Space, Toulouse, France

³ IRIT, University of Toulouse, CNRS, Toulouse INP, UT3, UT2J, Toulouse, France

THALES ALENIA SPACE OPEN

2 of 17

Table of contents

page 2

30/04/2025

1.

2.

3.

4.

5.

6.

Introduction

Dynamic mission planning problem

ATLAS2 – The AMAS approach

Benchmark scenarios and key performance indicators

Benchmark results

Conclusion & Perspectives

THALES ALENIA SPACE OPEN

3 of 17

Table of contents

page 3

30/04/2025

1.

2.

3.

4.

5.

6.

Introduction

Dynamic mission planning problem

ATLAS2 – The AMAS approach

Benchmark scenarios and key performance indicators

Benchmark results

Conclusion & Perspectives

THALES ALENIA SPACE OPEN

4 of 17

Introduction

page 4

30/04/2025

Agile Satellite Constellations

High Revisit frequency
Global coverage demand

Mission Planning Challenges

Multiple satellites
High variability in user needs
Time/resource constraints
Real-time adaptability

External evolutions: new requests arrival, change in weather forecasts, satellite or ground stations unavailability

Limitations of widely used Greedy Algorithm

Fast but input-order dependent
Poor adaptability to dynamic evolutions
Stop & Restart required for changes

🡺 Suboptimal in dynamic environments (Wang et al. 2018)

AI-Based Solution: Adaptive Multi-Agent System AMAS

Robust & responsive scheduling
Real-time request integration

🡺 Comparable or better than greedy (Bonnet 2017)

ATLAS2: Enhanced AMAS

ALL-IN-ONE © Thales Alenia Space

THALES ALENIA SPACE OPEN

5 of 17

Table of contents

page 5

30/04/2025

1.

2.

3.

4.

5.

6.

Introduction

Dynamic mission planning problem

ATLAS2 – The AMAS approach

Benchmark scenarios and key performance indicators

Benchmark results

Conclusion & Perspectives

THALES ALENIA SPACE OPEN

6 of 17

Dynamic mission planning problem

page 6

30/04/2025

Problem Definition

Set of agile satellites: S = {sat₁, sat₂, ..., satₙ}
Set of Candidate Acquisition Requests (CARs)

Each CAR has several Data Take Opportunities (DTOs)
CARs prioritized: Urgent > Nominal > Routine

Objective

Maximize the number of CARs assigned to an acquisition slot and affected to a satellite

🡺 Complexity

Agility of satellites increases placement flexibility and solution space
NP-Hard problem (Lenzen et al. 2011)

Traditional Methods

Small–scale systems: Linear/Constraint Programming (Lemaître et al. 2002)
Larger problems: Heuristics / Metaheuristics

Genetic Algorithms (Mansour & Dessouky 2010), Evolutionary Algorithms (Englander et al. 2012), Greedy Algorithms (Wang et al. 2011)

Limitations

Poor adaptability to dynamic/reactive contexts
Difficulty handling evolving constraints in real-time

🡺 Comparative analysis of methods (Lemaître 2002, Bonnet 2017, Wang 2020)

AMAS and HGreedy identified as best suited

Agility along pitch axis makes explode the combinatorial of the problem, due to the number of opportunities allowing

to acquire a single image

sat₁, sat₂, ..., satₙ

CAR Priority:

Urgent

Nominal

Routine

DTO duration

Satellite orbit

Area Of Interest

Max forward depointing

Max backward depointing

Null pitch

Acquisition slot

THALES ALENIA SPACE OPEN

7 of 17

Table of contents

page 7

30/04/2025

1.

2.

3.

4.

5.

6.

Introduction

Dynamic mission planning problem

ATLAS2 – The AMAS approach

Benchmark scenarios and key performance indicators

Benchmark results

Conclusion & Perspectives

THALES ALENIA SPACE OPEN

8 of 17

ATLAS2 – The AMAS Approach

page 8

30/04/2025

Agent Roles

Agents cooperate through local interactions to reach a global solution
Two roles of cooperative agents (AMAS4Opt model, Kaddoum 2011)

“Constrained” agents need help
“Service” agents offer help

Agent Types in Reactive Mission Planning

SAT Agent (Service role): Represents a satellite, evaluates CAR planning cost
CAR Agent (Constrained role): Represents a planning request, assesses its criticality
ACQ Agent (Mixed role): Represents Data Take Opportunities for a CAR on a satellite, helps CAR agents to be planned

🡺 Negotiation ensures dynamic conflict resolution between ACQ agents

AMAS Strengths

Adaptive Nature of AMAS

AMAS can operate continuously without interrupting real-time operations
Supports near real-time reactions to changes (e.g., weather updates)

Flexibility for Reactive Re-planning

Allows for dynamic updates to ready-to-send plans
Self-adaptive system handles disturbances naturally

Local interactions of agents leads to an emergent solution

AMAS Engine

🔄

Operates continuously
Uses idle time for optimization
Responds to updates

Conflict resolution via agent negotiation

♻

Ready-to-send plans

Disturbances

Plans

New Requests

THALES ALENIA SPACE OPEN

9 of 17

Table of contents

page 9

30/04/2025

1.

2.

3.

4.

5.

6.

Introduction

Dynamic mission planning problem

ATLAS2 – The AMAS approach

Benchmark scenarios and key performance indicators

Benchmark results

Conclusion & Perspectives

THALES ALENIA SPACE OPEN

10 of 17

Benchmark scenarios and key performance indicators��

page 10

30/04/2025

Evaluation Scenarios

Definition and Realism

Evaluation based on realistic user requests: spot, strip, mosaic, stereo
Constraints incorporated:

Satellite energy and memory limits
Real Guidance and attitude maneuver libraries

Scenario Structure

A scenario = set of Data Take Opportunities (DTOs) over a 24-hour window
Complexity managed by an index

Random priority assignment simulates user-defined criticality (CAR agent)

Scenario Set

4 scenario classes:

Increasing number of satellites and requests

3 complexity levels per class
Total of 12 scenarios ensure progressive and scalable evaluation

	Number of satellites	Number of requests	Scenario complexity index
Class	2	3 000	25	63	111
	5	6 000	29	65	110
	7	8 000	21	55	112
	10	12 000	21	60	127

Example of complexity index computation on a conflict horizon with 3 DTOs

Benchmark scenarios: satellites, requests and complexity

Densification of areas to increase scheduling conflicts

THALES ALENIA SPACE OPEN

11 of 17

Benchmark scenarios and key performance indicators��

page 11

30/04/2025

Ability of AMAS to continuously integrate high-priority requests

THALES ALENIA SPACE OPEN

12 of 17

Table of contents

page 12

30/04/2025

1.

2.

3.

4.

5.

6.

Introduction

Dynamic mission planning problem

ATLAS2 – The AMAS approach

Benchmark scenarios and key performance indicators

Benchmark results

Conclusion & Perspectives

THALES ALENIA SPACE OPEN

13 of 17

Benchmark results�

page 13

30/04/2025

Comparative Performance Analysis – AMAS vs HGreedy

Overview

HGreedy processes requests satellite-by-satellite, making non-revisable decisions, favoring speed over global optimization
AMAS allows cooperative reassessment of decisions, leading to more globally optimal solutions

Plan Score: Weighted by Priority

Performances of AMAS in delta with HGreedy score:

Key Insight: AMAS outperforms HGreedy in nearly all scenarios
Particularly effective in high-conflict, low-resource settings, e.g., Class 1 complex scenarios → +40% score improvement
Better satellite resource allocation due to global reasoning

Planned Acquisitions

Trend aligned with plan scores: More acquisitions completed by AMAS

Plans Score comparison - AMAS vs HGreedy

Number of planned acquisitions - AMAS vs HGreedy

THALES ALENIA SPACE OPEN

14 of 17

Benchmark results��

page 14

30/04/2025

Computation Time & Adaptability to Urgent Request

Computation Time

HGreedy: Rapid initial computation but inflexible
AMAS: Takes 2–3× more time initially (due to iterative reassessment), but remains within acceptable range for operational use
Valid trade-off when superior quality is required.

Adaptability to Urgent Request

Test Setup: Each scenario receives 10 random high-priority requests
HGreedy must restart planning → leads to time growth with size and complexity
AMAS adapts in < 1 min, even under increasing scenario complexity

Key Takeaways

AMAS excels in flexibility and real-time planning
Maintains coherent plans while prioritizing critical new demands
Enables mission-ready adaptability aligned with dynamic operational environments

Processing Time - AMAS vs HGreedy

Response Time to Urgent Requests - AMAS vs HGreedy

THALES ALENIA SPACE OPEN

15 of 17

Table of contents

page 15

30/04/2025

1.

2.

3.

4.

5.

6.

Introduction

Dynamic mission planning problem

ATLAS2 – The AMAS approach

Benchmark scenarios and key performance indicators

Benchmark results

Conclusion & Perspectives

THALES ALENIA SPACE OPEN

16 of 17

Conclusion & Perspectives�

page 16

30/04/2025

Advantages of the AMAS Approach

Distributed and Adaptive

Agents cooperate to make decentralized decisions
Enables real-time plan adjustments

Improved Resources Utilization

Better handling of conflicts and priorities
More efficient satellite tasking and responsiveness to urgent requests

Future Directions

Investigate hybrid solutions combining:

AMAS with heuristic strategies
Learning-based methods (e.g., RL, optimization via AI)

Tackle increasingly complex environments with scalable, intelligent planning architectures

Experimental results demonstrate that AMAS significantly outperforms the commonly used Hierarchical Greedy algorithm, especially in dynamic and reactive contexts

Mission planning

Scene acquisition

Image processing

On-board

Ground-based

Mission planning

IRMA project in IRT Saint Exupery:

Improve the responsiveness of Earth observation space systems

through a feedback loop from image analysis

to mission planning

THALES ALENIA SPACE OPEN

17 of 17

page 17

30/04/2025

Thank you for your attention.

Any questions?

THALES ALENIA SPACE OPEN