[agents] 2-year postdoc position at ONERA Toulouse on AI, Multi-Agent Systems and Optimization

[Gauthier Picard]

ONERA is opening a postdoc position for 2 years in the context of the 
European project DOMINO-E.

Title: Enabling Earth Observation Multi-Mission Dispatch and 
Communication Booking By Hybridizing Optimization and Machine Learning

Start of contract: 10/2023

Application deadline: 15/07/2023

Duration: 12 months, possibly extendable to 24 months

Net yearly salary: about 25 k€ (medical insurance included)

Keywords: Multi-agent resource allocation, planning & scheduling, online 
learning, reinforcement learning, multi-armed bandits, Earth observation 
satellite constellation, multi-mission

Profile and skills required: PhD in Computer Science, Artificial 
Intelligence or Operations Research with a strong publication record and 
a taste for theoretical and coding activities. Some prior knowledge in 
optimization, planning, scheduling, online learning, and reinforcement 
learning would be appreciated

Context:

In the context of the Horizon Europe DOMINO-E Innovation Action 
(https://domino-e.eu/), ONERA is involved in the development of novel 
techniques to demonstrate the feasibility of an innovative multi-mission 
federation layer for exploiting a set of space assets managed by various 
operators for various institutional and commercial applications. The 
goal of this federation layer is to efficiently use all these assets, so 
as to improve reactivity, persistence, precision and costs for various 
end-users. The federated layer will consist in smart services based on 
AI and machine learning, to be developed.

Proposed work:

ONERA is involved in two main scientific tracks: multi-mission coverage 
and dispatch, and multi-mission communication booking. The first track 
aims to decide how to dispatch observations of large areas to different 
constellations, instead of a single one, as to optimize some performance 
criteria, such as the make-span and the quality of the images. Indeed, 
in order to minimize the time to deliver images for a specific large 
area requiring multiple snapshots, the idea is to query several 
missions. However, since the missions may not be legacy (not owned by 
the system operator), some information such as the workload and the 
precise schedule of the satellites are not available. In order to divide 
the large area into sub-areas, and to allocate such sub-areas to 
multiple missions, this requires building/learning a workload model or a 
query acceptance model to guide the dispatch decisions. This learning 
problem is not straightforward, since the workload is both space and 
time dependent. Moreover, the large area coverage problem itself is also 
a hard problem, addressed in the literature using multi-satellite 
coverage of discrete points of a large area [1, 2], multi-satellite 
coverage using 2D-strips over a continuous polygon [3, 4, 5, 6], and 
mono-satellite or multi-satellite area scanning strategies [7, 8]. Yet, 
there is still a lot of place for optimizing large area splitting 
methods, to get a faster global area coverage. Some works also consider 
several coverage requests simultaneously [9, 10, 11, 12, 13, 14, 15], 
and define criteria to arbitrate their scheduling. However, no work take 
into account multi-satellite observations together with the management 
of the current load of each mission or urgent requests. Moreover, 
interfaces with external systems are not really discussed, and dynamic 
dispatch (dispatch step-by-step to different missions, management of the 
long-term impact of the ongoing dispatch decisions, management of 
uncertainties about the cloud cover, etc.) is still an open issue.
The second track aims to decide how to book communication stations, as 
to optimize the data freshness. This is based on the novel concept of 
GSaaS (Ground Station/Segment as a Service), where mission operators can 
make use of external ground stations to communicate with the satellites. 
Indeed, the current concepts of operations are mostly based on legacy 
networks of ground stations (either proprietary or long-term booking) 
with high trust and satisfaction rates. The idea of using other 
stations, proposed by GSaaS providers, is to reduce the time to access 
data on ground thanks to non-legacy stations, instead of waiting to get 
access to legacy stations which may be not frequently accessible. But, 
here again, the workload of these GSaaS services is not available, and 
thus building a workload model or query acceptance model of such 
stations is required to book the proper stations, at the best price. 
While the problem of booking communication slots exists in the 
literature [16], no approach takes advantage of the novel concept of GSaaS.
This post-doctorate is a real opportunity to develop strong research and 
apply it in the context of an innovating research project. This research 
will develop and evaluate AI-based and optimization techniques (such as 
multi-agent resource allocation, reinforcement learning, online 
learning, reasoning under uncertainties, decomposition methods, 
metaheuristics, etc.) to address these two tracks, and integrate them 
into the DOMINO-E modular architecture, in close interaction with Airbus 
Defense and Space, Cap Gemini and ITTI development teams, in the context 
of the Horizon Europe DOMINO-E project.

References:

     [1] Maillard, Adrien & Chien, Steve & Wells, Christopher. (2021). 
Planning the Coverage of Solar System Bodies Under Geometric 
Constraints. Journal of Aerospace Information Systems. 18. 1-18. 
10.2514/1.I010896.
     [2] Liu, Shufan & Hodgson, Michael. (2013). Optimizing large area 
coverage from multiple satellite-sensors. GIScience & Remote Sensing. 
50. 10.1080/15481603.2013.866782.
     [3] Niu, Xiaonan & Tang, Hong & Wu, L.. (2018). Satellite 
Scheduling of Large Areal Tasks for Rapid Response to Natural Disaster 
Using a Multi-Objective Genetic Algorithm. International Journal of 
Disaster Risk Reduction. 28. 10.1016/j.ijdrr.2018.02.013.
     [4] Ntagiou, Evridiki & Iacopino, Claudio & Policella, Nicola & 
Armellin, Roberto & Donati, Alessandro. (2018). Ant-based Mission 
Planning: Two Examples. 10.2514/6.2018-2498.
     [5] Chen, Yaxin & Xu, Miaozhong & Shen, Xin & Zhang, Guo & Zezhong, 
Lu & Xu, Junfei. (2020). A Multi-Objective Modeling Method of 
Multi-Satellite Imaging Task Planning for Large Regional Mapping. Remote 
Sensing. 12. 344. 10.3390/rs12030344.
     [6] Lenzen, Christoph and Dauth, Matthias and Fruth, Thomas and 
Petrak, Andreas and Gross, Elke Marie-Lena (2021) Planning Area Coverage 
with Low Priority. The 12th International Workshop on Planning & 
Scheduling for Space (IWPSS), 27-29. Jul. 2021.
     [7] Ji, Hao-ran & Huang, Di. (2019). A mission planning method for 
multi-satellite wide area observation. International Journal of Advanced 
Robotic Systems. 16. 172988141989071. 10.1177/1729881419890715.
     [8] Elly Shao, Amos Byon, Christopher Davies, Evan Davis, Russell 
Knight, Garrett Lewellen, Michael Trowbridge and Steve Chien (2018). 
Area Coverage Planning with 3-axis Steerable, 2D Framing Sensors. The 
28th International Conference on Automated Planning and Scheduling, June 
24–29, 2018, Delft, The Netherlands.
     [9] Lemaître, M., Verfaillie, G., Jouhaud, F. Lachiver, J.-M., and 
Bataille, N. (2002). Selecting and scheduling observations of agile 
satellites. Aerospace Science and Technology, 6(5):367–381.
     [10] Cordeau, J.-F. and Laporte, G. (2005). Maximizing the value of 
an Earth observation satellite orbit. Journal of the Operational 
Research Society, 56(8):962–968.
     [11] W ang, P., Reinelt, G., Gao, P., and Tan, Y. (2011). A model, 
a heuristic and a decision support system to solve the scheduling 
problem of an earth observing satellite constellation. Computers & 
Industrial Engineering, 61(2):322–335.s
     [12] Tangpattanakul, P., Jozefowiez, N., and Lopez, P. (2015). A 
multi-objective local search heuristic for scheduling Earth observations 
taken by an agile satellite. European Journal of Operations Research.
     [13] Zhu, W., Hu, X., Xia, W., and Sun, H. (2019). A three-phase 
solution method for the scheduling problem of using earth observation 
satellites to observe polygon requests. Computers & Industrial 
Engineering, 130:97–107.
     [14] Berger, J., Giasson, E., Florea, M., Harb, M., Teske, A., 
Petriu, E., Abielmona, R., Falcon, R., and Lo, N. (2018). A Graph-based 
Genetic Algorithm to Solve the Virtual Constellation Multi-Satellite 
Collection Scheduling Problem. In 2018 IEEE Congress on Evolutionary 
Computation (CEC), pages 1–10.
     [15] Zhibo, E., Shi, R., Gan, L., Baoyin, H., and Li, J. (2021). 
Multi-satellites imaging scheduling using individual reconfiguration 
based integer coding genetic algorithm. Acta Astronautica, 178:645–657.
     [16] A. Maillard, G. Verfaillie, C. Pralet, J. Jaubert, I. Sebbag, 
F. Fontanari, and J. Lhermitte . Adaptable Data Download Schedules for 
Agile Earth-Observing Satellites, Journal of Aerospace Information 
Systems 2016 13:3, 280-300

Host laboratory: ONERA, Toulouse, France

Applications including scientific CV, motivation letter, and letters 
from referees should be sent to Gauthier Picard 
(gauthier.picard at onera.fr) and Cédric Pralet (cedric.pralet at onera.fr)

-- 
Gauthier Picard, PhD, HDR
Directeur de Recherche / Senior Research Fellow
ONERA - DTIS - SYD
BP74025 - 2 avenue Edouard Belin, FR-31055 TOULOUSE CEDEX 4
Tel. +33 (0)5 62 25 26 54
https://www.onera.fr/en/staff/gauthier-picard/