Advances in space exploration require human presence in space for extended periods. Indeed, permanent settlement on other planets is now very much on the agenda, as are long space travels. However, supplying these missions from Earth would be too costly, if not impossible. A promising solution is to consider closed-loop systems and bioregenerative Life Support Systems (LSS) to ensure and maximize recycling of resources, in particular oxygen, water and food. Therefore, plant cultivation is an essential aspect that must be ensured while using resources as efficiently as possible.

In future missions, astronauts will have limited time for crop cultivation, as their main focus will be on completing mission objectives. Therefore, upcoming space cultivation systems must be designed to require the least crew intervention, incorporating more automation to streamline the process. Monitoring plant growth and health throughout their life cycle is crucial to ensure the proper operation of the closed-loop system. Advanced imaging techniques can gather essential data enabling non-invasive and automated assessments of the state of the plants and requiring minimal crew involvement. Swiftly identifying signs of nutrient deficiencies, drought or infections enables early response, and ultimately the success of long-term space missions.

In this context, the thesis aims to develop Artificial Intelligence (AI) tools to ensure the desired condition of the plants despite resource and environmental constraints. This challenge includes the optimization of plant cultivation systems, particularly in the extraction of useful information from their sensor measurements, as well as in the decision support and automation of these systems. It therefore focuses on AI for precision agriculture in LSS, with the overall objective of maximizing production, minimizing resource consumption, and more generally optimizing the system according to appropriate criteria (e.g. Advanced LSS Evaluator, ALiSSE). The thesis will delve into the two following interdependent challenges:

- estimating the state of plants and their environment using machine learning (ML) and computer vision (CV) algorithms, supported by recent datasets, pre-trained and foundation models,

- sequential decisions using planning and reinforcement learning (RL) algorithms to compute effective and economical autonomous cultivation strategies over the long term, benefiting from state-of-the-art plant cultivation models and simulators.

This study will pave the way for autonomous cultivation systems capable of analyzing and even reacting to the growing process in order to obtain plants in the required conditions while optimizing the use of space and resources (nutrients, water, energy, etc.).

Since 2019, more than twenty Master's students have contributed to research into precision agriculture for LSS during internships and research projects under the guidance of the supervisors of the thesis presented here. Progress on these projects was presented to the NASA KSC Space Crop Production team (Feb. 23), at MELiSSA conferences (Nov. 22 & Oct. 25) and at the UTIAS2 forum (Apr. 24). Test beds have been set up to collect data useful for the two main aims of the thesis, i.e. optimizing the perception capability and execution of these plant crop systems: a hydroponic system that can control both light and nutrient supply, and a Farmbot (open source agricultural robot). Both systems are equipped with sensors including a camera whose position can be controlled.

Regarding AI, this thesis tackles three main theoretical challenges:

- The control of a plant growth system through the estimation of the plant state (e.g. leaf area) induces a partially observable (PO) framework (e.g. PO Markov Decision Processes). This context poses the challenge of prior computation of estimators along with their conditional distributions. It also requires the development of optimization algorithms capable of handling the resulting complexity.

- Another challenge is the development of hybridization techniques between data-driven and model-based approaches to decision-making. On the one hand, robot control requires the use of classical, model-based planning, so that the computation of a solution is possible despite the large number of states. On the other hand, the various possible behaviors of the plants are not available in advance, and the related models must be learned through data collection with the plant growth system (e.g. Offline RL).

- The last challenge comes from the multi-criteria optimization (e.g. ALiSSE criteria): the objectives must be defined and aggregated appropriately to satisfy the requirements of the spatial context.

This thesis will result in publications in the AI fields of CV and sequential decision making under uncertainty, applied to space agriculture, as well as in the creation of datasets and algorithms enabling the refinement of autonomous plant growth systems for future space missions.

=================

For more Information about the topics and the co-financial partner (found by the lab!); contact Directeur de thèse - nicolas.drougard@isae-supaero.fr

Then, prepare a resume, a recent transcript and a reference letter from your M2 supervisor/ engineering school director and you will be ready to apply online  before March 13th, 2026 Midnight Paris time!