A personalized music-based neurotech solution has been recently deployed - BeatMove - allowing us to combat inactivity, accelerate rehabilitation, and help re-entering the virtuous cycle of physical and mental health and wellness. With this solution, individuals from all ages can break the vicious cycle of physical deconditioning, use our real-time, fun, personalized and adaptive music-based entrainment method of locomotion, increase physical activity, step by step, toward a healthier life. This internship will be a unique opportunity to (i) dive into the science of audiomotor synchronization, (ii) master modelling and real-time cueing techniques to deliver music-based rhythmic information delicately coupled to behaviour, (iii) conduct a lab-based experiment with TBD participants (healthy and/or patients), and (v) contribute to reach a global impact for our Health system in France, as moving-in-music can now be considered as an inexpensive and efficient non pharmacological intervention.

https://doi.org/10.1038/s41531-024-00852-6

https://doi.org/10.3389/fpsyg.2021.655121

Moving together in Augmented Reality

Benoît BARDY (EuroMov DHM) (with Julia AYACHE and Julien LAROCHE)

The SHARESPACE European project develops Shared Hybrid Spaces where human and artificial agents interact through fully embodied avatars in Extended Reality. Our goal is to foster immersive, meaningful social interactions and enrich user experience in digital spaces. In this context, the SYNC proof-of-principle investigates how synchronized movement enhances social connection in such spaces, and the role artificial agents can play in modulating group dynamics. After a phase of experimentation in Physical and Virtual Reality, this research is now extended to Augmented Reality (AR) environments. This internship will be a unique opportunity to (i) dive into the science of social sensorimotor synchronization and its role in social bonding, (ii) discover VR, XR and AR development tools, real-time motion capture techniques to analyse movement synchronisation, (iii) conduct and analyze experiments in AR settings to evaluate sensorimotor synchronisation, social connection, and their link, (iv) contribute to advancing XR technologies with potential applications in health, sport, and art sector.

https://doi.org/10.1145/3610661.3617989

https://doi.org/10.1038/s41598-020-74914-z

Modulating Motor Stereotypy through Auditory-Motor Synchronization

Loïc Damm (EuroMov DHM) (with Laurie Galvan and Sofiane Ramdani)

This interdisciplinary project investigates how auditory-motor synchronization (AMS) can modulate motor stereotypy—repetitive and rhythmical behaviors commonly observed in neurodevelopmental disorders such as autism. Grounded in sensorimotor integration theory and neurophysiological insights into parvalbumin (PV) interneuron dysfunction, the study explores whether AMS training can alter gamma-band oscillations (GBO) and promote motor repertoire enrichment. The internship will offer a unique opportunity to:

(i) explore mechanisms underlying stereotyped behaviors and their cortical correlates;

(ii) engage in human-subjects experimentation with EEG and motion capture tools;

(iii) analyze complex data (kinematics, spectral EEG, nonlinear measures);

(iv) contribute to the development of therapeutic strategies targeting sensorimotor plasticity.

The project bridges neuroscience, digital health, and embodied interaction, with potential applications in clinical interventions for ASD and related conditions.

https://doi.org/10.1177/13623613221105479

https://doi.org/10.1093/pnasnexus/pgae132

Influence of auditory stimulation on locomotor resilience

Loïc Damm (EuroMov DHM) (with Antoine Dufourneau and Nicolas Sutton Charani)

Walking in natural environments requires balance control to prevent falls, especially in older adults. Resilience, the ability to resist and recover from disturbances, is key but understudied in this population. This study examines how different rhythmic auditory cues (adaptive, 1/f, periodic, randomized) affect elderly participants’ balance during walking in a disturbed environment. Findings may highlight the potential of auditory cues for enhancing balance control.

https://doi.org/10.1242/jeb.237073

https://doi.org/10.1080/14763141.2023.2298958

Gesture-speech synchronization in dyad

Ludovic MARIN (EuroMov DHM) (with Patrice GUYOT)

The SYNCOGEST project is an ANR based funding aiming to model spontaneous gestures in face-to-face interactions to enhance the naturalness of embodied conversational agents. The master project will be part of the Syncogest consortium and will conduct a multimodal data collection (audio, video, MOCAP) of a dyadic corpus to analyze gesture-speech synchronizations. Twenty dyads will be scrutinized when interacting together in an open discussion. The goal of the master will be to capture and analyze gestures and body motions that are relevant for nonverbal communication. The gestures will be first annotated by linguists using the method of Rohrer et al. (2023), while the master will analyze the kinematic properties (velocity, amplitude, etc.) of each dyad. Hence, in a second part, the master will help building along with experts in AI, the automation of gesture segmentation using the method of Lozano-Goupil et al. (2022) in order to develop a general methodology of extraction of relevant movements for any dyadic conversations. This work will contribute to improving the modeling of human-like co-speech gestures in conversational AI

https://doi.org/10.1177/25152459221140842

https://doi.org/10.1016/j.neuropsychologia.2022.108347

Towards a User-friendly Agent-based Model of Approach and Avoidance Dynamics Regarding Physical Activity for Health

Christophe Gernigon (EuroMov DHM) (with Sylvain Vauttier and Rémi Altamore)

For people who have been physically inactive for a long time, regular physical activity goals can be either rewarding challenges to be met, or threats to self-esteem to be avoided by all manner of excuses. The dynamics of approach and avoidance motivations that result from such perceptions have just been modeled as agent-based models (ABM). However, this ABM is still predominantly formal and has only been validated in a competitive sport context. The aim of this master's internship will be to develop a user-friendly Netlogo-type (http://ccl.northwestern.edu/netlogo) computer version of the ABM, and to validate this version by comparing the statistical properties of its simulation outputs with those of longitudinal data from people who need to engage in regular physical activity for their health.

References:

https://doi.org/10.1037/gpr0000055

https://doi.org/10.1177/1948550617691100

Towards a User-friendly Agent-based Model of Approach and Avoidance Dynamics in Sport

Christophe Gernigon (EuroMov DHM) (with Sylvain Vauttier and Rémi Altamore)

Competition athletes progress in their careers by pursuing ambitious goals, which can either be rewarding challenges to overcome, or threats to self-esteem that can lead them to relinquish their goals. The dynamics of approach and avoidance motivations that result from such perceptions have just been modeled as agent-based models (ABM). However, this ABM is still predominantly formal. The aim of this master's internship will be to develop a user-friendly Netlogo-type (http://ccl.northwestern.edu/netlogo) computer version of the ABM, and to validate this version by comparing the statistical properties of its simulation outputs with those of longitudinal data from athletes pursuing a goal. Ultimately, the model built is intended to be used by coaches and athletes to virtually test the effects of mental preparation interventions according to their timing and duration.

References:

https://doi.org/10.1037/gpr0000055

https://doi.org/10.1177/1948550617691100

Muscle parameter estimation: comparison between healthy and post stroke subjects

François Bailly, Pierre Schegg (CAMIN, Inria Montpellier)

During this internship, the student will utilize the U-Limb dataset [1] to develop realistic musculoskeletal simulations of daily life activities for both healthy individuals and post-stroke patients. These simulations will be implemented using BioRBD and Bioptim software. The student will then compare various algorithms (e.g. [2]) to accurately estimate the muscle parameters (maximal isometric force, optimal fiber length, pennation angle, tendon slack length and maximal contraction velocity) of the model. Different methods and algorithms will be compared to determine their accuracy and effectiveness. Additionally, the student will investigate which movement characteristics lead to accurate parameter estimation and good convergence properties. The internship will also involve studying post-stroke movement modeling and comparing parameter calibration techniques and results between healthy subjects and post-stroke patients to identify key differences.

https://doi.org/10.1093/gigascience/giab043

10.1016/j.jbiomech.2015.11.006

Development of a Human Machine Interface for safe and efficient Human-Robot experimentation

François Bailly, Pierre Schegg (CAMIN, Inria Montpellier)

In this project, the student will assist in designing and implementing software features for experiments involving human subjects interacting with manipulator-type robots.
The student will review Human-Robot Interfaces used in previous similar experiments [1] and User Experience (UX) guidelines to design and implement visualization tools. These tools will be used before the experiment to explain the expected trajectory to participants and as a tracking tool during the experiment.
Additionally, the student will develop software tools to enable engineers and researchers to monitor the experiment's progress and display key metrics, as well as perform a posteriori data analysis [2].
The student may also contribute to implementing safety features and software safeguards to ensure safe interaction between the robot and the participants.

References

https://doi.org/10.1101/2024.01.11.575155

10.1016/j.compbiomed.2024.109434

Characterizing the microstructure of reaching movements after a stroke for better rehabilitation

Denis MOTTET, Karima BAKHTI and Emmanuel GUIGON

(EuroMov DHM, CHU Montpellier, ISIR Paris)

Rehabilitation is key to sensorimotor recovery after stroke. Yet, it is still unclear how to tailor the nature and the dose of the therapy to the exact functional deficits of each individual patient.

Does characterizing the microstructure of reaching movements for each patient help to better identify his/her specific deficit to improve his/her rehabilitation?

In this internship, we will monitor the microstructure of reaching movements in people with stroke and healthy controls, and interpret the observed deficits with a biologically plausible computational model of sensorimotor control calibrated for each patient.

References:

https://doi.org/10.1016/j.apmr.2013.10.006

https://elifesciences.org/articles/88591

https://hal.archives-ouvertes.fr/hal-03276320v2

ReArm AI Coach to Monitor and Increase Paretic Arm Use in Daily Activities of Stroke Patients

Karima BAKHTI (EuroMov DHM) (with Makii MUTHALIB and Abdellak IMOUSSATEN),

Monitoring and enhancing paretic arm use in stroke patients through AI-driven

personalized rehabilitation strategies.

This project aims to develop a virtual coach to improve recovery after strokes with the

potential to extend its application to other disorders. By leveraging clinical trials, wearable

sensors, and artificial intelligence, the virtual coach will monitor arm use after a stroke

and provide real-time feedback to patients and therapists. The goal is to break the vicious

cycle of paretic arm non-use and improve functional recovery, ultimately enhancing the

quality of life for stroke survivors.

References

https://doi.org/10.1002/ana.25679

https://doi.org/10.1186/s13063-021-05689-5

Clinical Validation of Visuomotor Performance Metrics in Post-Stroke Rehabilitation

Karima BAKHTI (EuroMov DHM) (with Makii MUTHALIB et Denis MOTTET)

This project aims to evaluate upper limb performance markers, such as speed and precision, during functional daily movements in post-stroke individuals. We will analyze the kinematics of visuomotor tasks involving a speed-accuracy trade-off (circular steering) with continuous feedback. Measurements have been conducted using a graphic tablet and a Kinect sensor. The goal is to assess the feasibility and relevance of these tools in a clinical setting. A comparison will be made between the performances of post-stroke patients and healthy subjects. We aim to validate the sensitivity of this instrumented evaluation against clinical scores, particularly the Wolf Motor Function Test and The Block and Box Test. This approach will help better quantify motor deficits and rehabilitation progress. It will contribute to refining physiotherapy management. The protocol will include statistical analyses of correlation and discrimination. This internship is part of a translational effort toward clinical practice.

Références

doi: 10.1177/15459683251331582

https://doi.org/10.1186/s13063-021-05689-5

Predicting Real-World Arm Use After Stroke: From Clinical Data to Smart Models

Karima BAKHTI (EuroMov DHM) (with Makii MUTHALIB and Nicolas SUTTON-CHARANI)

This internship focuses on developing predictive models of real-world arm use in daily life among post-stroke patients, using already collected clinical and contextual data. Arm use is measured via accelerometry (funcUseRatio), while predictors include motor scores (Fugl-Meyer), functional tests (Wolf Motor Function Test, Box and Block Test), autonomy (Barthel Index), and a semi-structured interview exploring barriers to upper limb use. The intern will apply supervised learning methods to train models on complete datasets and generate predictions in cases where accelerometric data are unavailable. The main goal is to identify predictors associated with non-use of the paretic upper limb in daily activities. This approach will help reveal discrepancies between expected and actual use, shedding light on behavioral or environmental factors. The project aims to refine personalized rehabilitation strategies while accounting for uncertainty in the data.

Références :

doi : 10.1109/ICMLA.2013.26

https://doi.org/10.1186/s13063-021-05689-5

Machine Learning for Personalized HD-tDCS Dosage in Stroke Rehabilitation

Makii MUTHALIB, Karima BAKHTI, Abdellak IMOUSSATEN, Denis MOTTET (EuroMov DHM)

Standardized “one-size-fits-all” protocols for transcranial direct current stimulation (tDCS) often overlook individual variability in stroke rehabilitation. The OptiStim project has already demonstrated using machine learning models that functional near-infrared spectroscopy (fNIRS), specifically its cerebrovascular and cardiovascular components, captures individualized physiological signatures relevant for personalizing neuromodulation. Utilizing fNIRS data from the ReArm clinical trial, this internship will focus on optimizing and enhancing this existing machine learning pipeline. This involves deep dives into advanced signal decomposition methods, innovative feature engineering, and rigorous model refinement to accurately predict individualized HD-tDCS responses. The optimized predictive model will undergo stringent validation against objective clinical recovery metrics, such as accelerometry-based paretic arm use and standardized motor scores. The long-term objective is to enable real-time, patient-specific stimulation dosing, thereby laying the foundation for sophisticated adaptive tDCS systems that can significantly improve functional recovery in stroke patients.

https://doi.org/10.1111/ner.12632

https://doi.org/10.1186/s13063-021-05689-5

Averted gaze and social anxiety in patient-healthy participants interaction

Ludovic MARIN (EuroMov DHM) (with Mathilde PARISI and Stephane RAFFARD)

The Enhancer project is an ANR based funding aiming to understand the speech/gesture interactions during a social conversation between a patient suffering from schizophrenia and a healthy participant. Schizophrenia is a complex psychiatric condition marked by a heterogeneous clinical profile, in which impairments in social interaction are a core feature. Individuals with schizophrenia often display atypical verbal and nonverbal behaviors during social exchanges. The master project will be part of the Enhancer consortium and will investigate the contributions of averted gaze and social anxiety to these interpersonal communication deficits. To achieve this, the master will employ automated gaze detection using OpenFace and OpenGaze, alongside gesture analysis derived from video recordings processed with MediaPipe in order to compare the nature of the interaction deficits.

References

Parisi, M., Raffard, S., Fauviaux, T., Vattier, V., Mrabet, D., Capdevielle, D., & Marin, L. (In press). Emotional Mimicry and Smiling Behaviors in Schizophrenia: An Ecological Approach. Schizophrenia

Parisi, M., Raffard, S., Slangen, P., Kastendieck, T., Hess, U., Mauersberger, H., Fauviaux, T., & Marin, L. (2024). Putting a Label on Someone: Impact of Schizophrenia Stigma on Emotional Mimicry, Liking, and Interpersonal Closeness. Cognition and Emotion, 1–17. https://doi.org/10.1080/02699931.2024.2339531

Evidential fusion of actimetric data coming from several sensors

Nicolas Sutton-Charani (EuroMov DHM)

The theory of belief function proposes a formal framework where uncertainty can be modelled with different levels leveraging the adaptation of uncertain models to the uncertainty type (aleatory or epistemic). In this framework several combination rules have been proposed that can imply the fusion of the information collected through different sensors. This internship will compare different evidential fusion strategies (concatenation, conjunctive, disjunctive) inside ML models in the activity recognition perspective.

References

Wenjun Ma, Yuncheng Jiang, Xudong Luo, A flexible rule for evidential combination in Dempster–Shafer theory of evidence, Applied Soft Computing, Volume 85, 2019, 105512, ISSN 1568-4946, https://doi.org/10.1016/j.asoc.2019.105512.

Shafer, Glenn. A Mathematical Theory of Evidence. Princeton University Press, 1976. JSTOR, https://doi.org/10.2307/j.ctv10vm1qb.

Rhythmic Auditory Stimulation (RAS) against Neurodegenerative Diseases: A Multiscale Modeling Approach

Mathieu DESROCHES [Inria, MathNeuro project-team] (with Benoît BARDY, Fabien CAMPILLO and Loïc DAMM)

RAS is currently being tested as a rehabilitation strategy for people suffering from Parkinson's Disease (PD) [1]. Through this internship, we will develop a model of this protocol using the framework of coordination dynamics. Namely, we will inspire from the Haken-Kelso-Bunz (HKB) model [2], with a time-dependent forcing signal mimicking the RAS. We will study the dynamic states that such a model is capable of producing, focusing on obtaining a stabilizing effect of the RAS, deriving a map of the parameter space. Then, we will calibrate the model with real data from BeatMove [3] in order to assess the parameter region where movement stabilization can be obtained. In a parallel approach, we will investigate the potential higher-order interaction effect of the RAS [4]. To this end, we will consider a coupled hybrid oscillator of van der Pol / Rayleigh [2] (from which the HKB model was derived) for limb coordination, and apply the RAS on the link between each node/oscillator, hence representing a time-periodic modulation of the connection between the two limbs. Similar to the previous approach, we will calibrate this coupled model and map its parameter space in order to unveil how to best enhance movement stability.

[1] https://doi.org/10.1038/s41531-024-00852-6

[2] https://doi.org/10.1007/s00422-021-00890-w

[3] https://www.beatmove.fr/en/

[4] https://doi.org/10.1038/s41593-022-01070-0

Mutual Impact of Hip and Knee Joint configurations on Torque production

François Bailly and Christine Azevedo (INRIA, CAMIN)

While studies of the human movement often focus on joint torques and kinematics, many isolate single joints, overlooking multi-joint interactions. In this internship, we want to explore the impact of hip joint configurations on the knee and conversely. For that purpose, we already acquired data from fourteen participants (7 females, 7 males) performing maximal voluntary contractions on a dynamometer with their right hip and knee. Four types of acquisitions were performed: passive, isometric, isokinetic concentric and isokinetic eccentric acquisitions for several hip and knee angle configurations. The goal of this internship is to organize and preprocess raw data collected during biomechanical experiments, develop scripts or pipelines for data cleaning and normalization, create insightful visualizations to represent joint kinetics/kinematics and compare to literature data. Finally, we expect the intern to assist in the biomechanical interpretation of findings.

https://doi.org/10.1016/j.jbiomech.2007.03.022

https://doi.org/10.1186/s40798-021-00330-w

Video Analysis and Artificial Intelligence for Early Detection of Writing Disorders in Children

Binbin Xu, Frédéric Puyjarinet and Gérard Dray (EuroMov DHM)

Contact : gerard.dray@umontpellier.fr

Systems for detecting fine motor skills disorders traditionally rely on instrumented sensors (tablets, pencils, accelerometers). More recently, markerless approaches using computer vision have emerged, thanks in particular to advances in pose estimation algorithms such as OpenPose and MediaPipe. These methods make it possible to capture arm, hand, and facial movements from simple videos, paving the way for non-invasive and accessible assessments. Initial proof-of-concept studies show an ability to detect signs of dysgraphia or developmental coordination disorder (DCD) with encouraging accuracy. Edge AI systems are particularly well suited to these contexts, as they ensure local data processing, respecting confidentiality constraints. However, the robustness of detections, multimodal synchronization (movement, posture, expressions), and model interpretability remain major challenges.

As part of this internship, you will draw up specifications and conduct initial tests for a proof-of-concept device for detecting writing disorders in children.

Maggioni, V.; Azevedo-Coste, C.; Durand, S.; Bailly, F. Optimisation and Comparison of Markerless and Marker-Based Motion Capture Methods for Hand and Finger Movement Analysis. Sensors 2025, 25, 1079. https://doi.org/10.3390/s25041079

Rangasrinivasan, S., Sumi Suresh, M.S., Olszewski, A. et al. AI-Enhanced Child Handwriting Analysis: A Framework for the Early Screening of Dyslexia and Dysgraphia. SN COMPUT. SCI. 6, 399 (2025). https://doi.org/10.1007/s42979-025-03927-0

Synchronized EEG-fNIRS for Enhanced Brain Source Validation in Real-World Environments

Stéphane Perrey (EuroMov DHM, UM), Binbin Xu (EuroMov DHM IMT), Gérard Dray (EuroMov DHM, IMT), Jochen Baumeister (Univ Parderborn, Germany)

Electroencephalography (EEG) offers high temporal resolution for monitoring brain activity, and with recent technological advances, it is now viable in real-world, mobile contexts. This shift from lab-based studies to ecological environments is crucial for applied fields such as sports and occupational neuroscience, enabling more accurate assessments of brain function during natural movement. Functional Near-Infrared Spectroscopy (fNIRS) measures cortical hemodynamics with high spatial precision and is similarly portable and non-invasive. Combining EEG and fNIRS produces rich, complementary datasets that are ideal for machine learning. ML techniques can improve source localization and pattern recognition, enabling deeper insights into cognitive states than either modality alone. This integration enables advanced predictive modeling of brain activity, particularly in complex and dynamic environments.

A multimodal EEG-fNIRS system powered by ML can improve brain monitoring and performance optimization in sports and exercise by enabling real-time, personalized feedback. Beyond sports, this approach has broader applications in health and rehabilitation, where understanding of brain measures (e.g. load, fatigue) is critical.

Research Question

How can the integration of mobile EEG and fNIRS, enhanced by machine learning, improve the accuracy and ecological validity of brain source localization and brain measures/biomarker classification during real-world sports performance?

1 Bunterngchit C, Wang J, Hou ZG. Simultaneous EEG-fNIRS Data Classification Through Selective Channel Representation and Spectrogram Imaging. IEEE J Transl Eng Health Med. 2024 Aug 23;12:600-612. doi: 10.1109/JTEHM.2024.3448457.

2. Phukhachee T, Angsuwatanakul T, Iramina K, Kaewkamnerdpong B. A simultaneous EEG-fNIRS dataset of the visual cognitive motivation study in healthy adults. Data Brief. 2024 Feb 27;53:110260. doi: 10.1016/j.dib.2024.110260.

Modelizing the recurrency of a new physical activity habit assessed by accelerometer

Julie Boiché (with Emmanuel Le Clezio and Rémy Dadier)

Euromov DHM - IES

Julie.boiche [@] umontpellier.fr

While controlled psychological processes are deemed essential to drive the initiation of structured and newly adopted behaviors, automatic factors play a major role in their maintenance through time. There are few longitudinal studies reporting data in individuals who regularly adopt a new physical activity habit, and all of them used self-report tools to assess automatic processes (i.e., habits scores) and behavior adoption. The purpose of the internship will be to (1) recruit a sample of healthy individuals that intent to adopt a new active routine on a regular basis and will be equipped with a GT3X accelerometer during a 4-week follow-up ; (2) use classification techniques to track the adoption of the new PA behavior and (3) conduct time-series analyses to estimate the increase of objectively assessed behavior adoption through time.

The internship will be supervised by a mixed team from the Euromov and IES laboratories, respectively specialists in human movement and the analysis of temporal signals.

References

https://doi.org/10.1016/j.psychsport.2018.12.007

https://doi.org/10.1111/sms.12730

doi:10.51257/a-v1-r1815

doi : 10.51257/a-v1-te5220

https://www.machinelearningplus.com/time-series/time-series-analysis-python/

Evaluation of levodopa responsiveness using actimetry in Parkinson’s disease

Cécile Aerts, Alice Bourdon (Clinique Beau Soleil), Nicolas Sutton-Charanni and Denis Mottet (Euromov DHM)

Parkinson’s disease is the second most common neurodegenerative disorder. When oral medication is no longer sufficient to control symptoms, continuous dopaminergic stimulation (CDS) can be considered. Among these second-line treatments are drug infusion pumps and deep brain stimulation (DBS). Their implementation requires a prior assessment of the patient’s levodopa responsiveness, which is currently based on an in-hospital levodopa challenge test. This test is burdensome and sometimes difficult for patients to tolerate. The DopActi study aims to develop a simpler alternative using actimetry, a technology that records patients’ movements in their daily environment. The collected data will be analyzed with artificial intelligence to assess levodopa responsiveness. This pilot study, conducted in Montpellier, brings together clinicians, movement analysis researchers, and an industrial partner providing the actimetry devices.

https://doi.org/10.3389/fneur.2023.1080752

https://doi.org/10.1007/s00415-020-09810-7

Title

Supervision

10-line abstract

2 refs