List of winning projects
PhD thesis
Project title: Constraint-based control of an over-actuated agricultural robotic system, to coordinate the carrier and the implement around a precise agronomic point of action, in a resource-efficient manner that minimises risks to crops.
This PhD thesis aims to develop a generic framework for the control of over-actuated agricultural robotic systems dedicated to precision agronomic intervention. New agroecological practices require localised actions as close as possible to crops, weeds or the soil, for which the quality of the intervention no longer depends solely on the robot’s trajectory, but on the actual movement of the agronomic point of action. However, modern agricultural machinery may possess additional degrees of freedom, such as lateral offset, orientation, retraction or height adjustment. These degrees of freedom introduce redundancy in actuation: a single task can be performed through various combinations of the carrier’s movement and the tool’s movement. The primary objective of the thesis will be to characterise the space locally accessible to the agronomic point of action, taking into account the system’s kinematic, dynamic and physical constraints. Not all geometrically similar movements are necessarily feasible, nor are they equivalent in terms of energy consumption, risk to crops or quality of intervention. The thesis will therefore seek to define metrics suited to this space, in order to compare feasible solutions beyond mere geometric error. These metrics will then be used to develop control and optimisation laws capable of exploiting the redundancy between the carrier and the tool, ranging from reactive approaches to predictive formulations. The proposed methods will be validated experimentally on several robotic platforms and in localised intervention scenarios, with mechanical weeding as the primary application. The trials will notably involve the TSCF platforms, the Montoldre plots, the Roboterrium and the pixel cropping scenarios associated with the NinSAR project. This validation will enable an assessment of the proposed framework’s generalisability across different types of carriers, tools and modes of locomotion, as well as its performance in terms of point-of-action accuracy, behaviour under conditions of uncertainty, adherence to safety margins relative to crops, energy cost and the quality of agronomic intervention.
Postdoctoral Research
Project title: Disturbances induced by agricultural inputs on genotype–soil–microbiome interactions in wheat and the creation of a platform to assess the impact of the products used on microbial diversity.
- Postdoctoral researcher: María Ariza
- Affiliated unit: UMR 1488 – RIBP (Résistance induite et de bioprotection des plantes)
- Co-supervisors: Angélique RAT (UMR 1488 RIBP, University of Reims Champagne-Ardenne) and Leonardo BRENNER (LICIIS - CEA UR 3690 LRC DIGIT, University of Reims Champagne-Ardenne)
- Project duration: 2026–2028
PREDIBLÉ is an interdisciplinary project at the interface between digital agroecology, microbial ecology and the modelling of complex systems. The project aims to predict the effects of agricultural inputs (pesticides, fertilisers, biocontrols and biostimulants) on wheat-associated microbiomes by integrating the interactions between soils, plant genotypes and microbial communities. PREDIBLÉ is based on four complementary work packages: (i) the construction of a harmonised database combining microbiome sequencing data, environmental metadata and farming practices; (ii) the development of integrative models combining statistical learning, Bayesian networks and formal modelling (finite state machines and Petri nets); (iii) the experimental validation of the models using field plots; and (iv) the creation of a digital decision-support platform to simulate the impact of agricultural practices on the diversity and stability of wheat microbiomes.
Project title: ORIGAMI: Organise, Represent, Implement – co-designing agroecological systems through innovative digital mediations.
- Postdoctoral researcher: Aliénor Morvan
- Affiliated unit: Centre for Design Research (CRD, ENSCI – ENS Paris Saclay) at the École Nationale Supérieure de Création Industrielle (ENSCI)
- Co-supervisors: Lorène Prost (UMR 1048 SADAPT, INRAE) and Olivier Hirt (Centre for Design Research (CRD), ENSC)
- Project duration: 2026–2028
The ORIGAMI project analyses the use of digital tools (collaborative platforms, modelling, visualisation) in regional intermediation to support the agroecological transition (TAE). It explores three questions:
- Organising: How do these tools foster new forms of cooperation and bring together a community committed to the TAE?
- Representation: How do they enable the complexity of agroecological systems to be represented in a way that makes it more accessible?
- Implementation: How do they support the transfer of practices that can be adapted to different contexts?
It aims to make scientific, methodological and operational contributions. The project will formalise a framework for analysing socio-technical mediation practices and a typology of the functions of digital tools in territorial co-design initiatives. It will produce mock-ups, and even prototypes of tools, visualisation systems and methodological resources (guides, feedback reports), all made available as open source. The project aims to help shape an interdisciplinary field spanning agroecology, design, human-computer interaction (HCI), and the social and cognitive sciences, by studying the role of designers and the interplay between facilitation strategies, co-design and the adoption of tools by intermediary actors and stakeholders. Three case studies, based on existing fieldwork with established partnerships (PNR du Pilat, MAELIA, Colza Story), will identify replicable approaches across different regions, peer farmers and intermediary teams, to strengthen their capacity to design and use appropriate tools. The project will examine the conditions for equitable participation by reducing power imbalances and asymmetries. It will highlight the importance of coordination for TAE whilst incorporating a frugal approach into the design of the tools to limit their environmental impact.
Project title: Development of a foundational model for processing multimodal remote sensing data (hyperspectral, lidar, etc.) and improvement of decision-support algorithms (enhanced performance, reduced complexity) in agroecology.
- Postdoctoral researcher: Léo Paillet
- Affiliated unit: Applied Mathematics and Computer Science, Paris Saclay
- Co-supervisors: Romain Thoreau (UMR MIA, AgroParisTech) and Jean-Baptiste Féret (UMR TETIS, INRAE)
- Project duration: 2026–2028
Remote sensing provides large quantities of multimodal data (optical, lidar, radar), enabling the assessment of production, regulation and the support of agroforestry ecosystems and agroecological infrastructure. This data remains largely underutilised, whether due to the complexity of using several different types of data simultaneously or a lack of manual annotation. Furthermore, decision-support models are generally designed for a single application, for one type of data. We propose a general approach to utilising these data by investigating self-supervised learning methods for multimodal representations, using foundation models, within the context of agroecology. These models can extract latent features that summarise the information and are useful for a wide variety of tasks, without requiring annotation for their training. These representations are then utilised by more specific decision-support models. We propose to evaluate current methods and their limitations using synthetic data generated by the DART physical model, then improve them, and finally apply them to real-world data. Indeed, this postdoctoral project lies at the interface of two projects funded by the PEPR Agroecology and Digital Technology programme: MELICERTES (modelling and assessment of organic carbon content using multimodal data) and TwinFarms (development of digital twins for decision support, using multimodal satellite data). The methodological developments of this postdoctoral project will therefore be assessed in light of the objectives of these two projects. Thus, our interdisciplinary approach will enable better use to be made of the vast amount of data collected in agroecology and will improve the performance of current decision-support models, with a practical application.
Project title: Capitalising on and harnessing knowledge about Diversified Agrosystems, by utilising text mining methods applied to heterogeneous text corpora (social media, press articles, technical bulletins, etc.).
- Postdoctoral researcher: Oussama Mechhour
- Affiliated unit: UMR 1248 AGIR (AGroécologie - Innovations - TeRritoires)
- Co-supervision: Sandrine Auzoux (UR AIDA TA B-115, CIRAD) and Noémie Gaudio (UMR AGIR, INRAE)
- Project duration: 2026–2028
A key driver of the agroecological transition lies in the diversification of agrosystems, particularly through intercropping, which allows for the exploitation of complementarities between species and reduces dependence on synthetic inputs. Agronomic trials provide valuable quantitative data on these systems. However, they remain limited in the number of species studied and often focus on yield, to the detriment of other ecosystem services. At the same time, a wide range of practical knowledge exists amongst those working on the ground (farmers, advisers, seed producers). However, this knowledge is scattered across a variety of heterogeneous, unstructured and highly context-specific textual sources (social media, local press, technical bulletins, etc.). The aim of this postdoctoral research is to develop a system for consolidating and harnessing knowledge on diversified agrosystems, capable of automatically extracting and structuring, from a variety of textual corpora, information relating to crop diversification practices (soil and climate contexts, implementation methods, observed performance and feedback). To this end, natural language processing methods will be employed to transform this heterogeneous textual data into structured, traceable and interoperable knowledge. Among the major scientific challenges to be overcome are the adaptation of general language models to a highly specific agricultural domain, rich in technical terminology and contextual variations, the limited availability of annotated data representative of crop diversification practices, which necessitates the development of domain-specific data augmentation strategies capable of enriching training datasets without compromising the agronomic consistency of the generated information; and the frugality of the solutions developed (computational efficiency, reduction of computational costs and the environmental footprint – particularly carbon emissions – and the ability to operate in resource-constrained environments). These methods will enable quantitative approaches to be complemented by expert knowledge, thereby grounding the understanding of agrosystems more firmly in agricultural reality. This project also incorporates a responsible digital approach, evaluating the solutions developed according to their effectiveness, computational cost and ability to function in resource-constrained environments.
Project title: Mobilising digital tools to predict the risk of ground vole outbreaks in grassland livestock systems and to support collective, preventive and effective agroecological management.
- Postdoctoral researcher: Hélène Lisse
- Affiliated unit: UR 1471 – TSCF (Technologies et systèmes d’information pour les agrosystèmes)
- Co-supervisors: Adrien Pinot (USC1233 RS2GP, VetAgroSup) and Riccardo Bertoglio (UR1471 TSCF, INRAE)
- Project duration: 2026–2028
The project aims to develop digital tools to improve the monitoring and preventive management of ground vole outbreaks in mid-mountain grassland systems (Massif Central, Jura). These outbreaks involve densities of up to 500 adult voles per hectare of grassland, causing considerable losses of forage resources. In peak years, which can occur every 5 to 8 years, high vole densities cause lasting damage to large areas of permanent grassland, leading to direct forage losses as well as significant collateral damage that undermines the self-sufficiency of farms (losses estimated at 10,000 euros EBE/UTH). The proposed programme addresses two major operational challenges in the management of this grassland pest: the difficulty of detecting low-density outbreaks at an early stage across large areas, and the lack of action areas based on landscape units that are genuinely relevant to population dynamics. The project will combine satellite remote sensing, parcel-level data, field observations (vole density data and drone imagery). It will enable the identification of functional landscape units, the production of risk maps at the plot level, and the development of an AI algorithm to detect mounds (soil heaped up by voles) at an early stage using drone orthophotos. These components will be integrated into a prototype decision-support tool designed to target surveys, prioritise areas for intervention and support earlier, more targeted and more collective agroecological control measures.
