Jobs

Context and Scientific Positioning

Walking is a fundamental motor activity at the heart of sustainable mobility, accessibility, and public health issues. However, the walking models used today in biomechanics, humanoid robotics, and pedestrian navigation are mostly based on average and normative representations of movement. These approaches describe the kinematics of walking in controlled environments, largely neglecting interindividual variability and the subtle influence of the urban context on locomotor dynamics.

Within the GEOLOC laboratory (Gustave Eiffel University), several studies have contributed to moving beyond this standardized view of walking by developing high-precision pedestrian positioning solutions based on embedded inertial sensors. Research conducted as part of the ANR-DGA MALIN and H2020 ICHASE projects has led to the development of robust systems for tracking movement in constrained environments. More recently, the ANR INMOB LabCom (2021–2025) introduced an approach based on personalized walking models, using machine learning to construct individual "walking fingerprints" from inertial data collected in real-world conditions.

This work has shown that walking signatures are highly specific to each individual, that they are stable over time, and that they are sensitive to environmental conditions and sensor positioning.
However, while this research has improved pedestrian navigation and individual movement modeling, it has not yet systematically integrated urban morphology and environmental factors as explanatory variables of walking dynamics.

The proposed thesis is part of the ANR CITY-STEP project, which aims to renew the analysis of walkability by combining detailed biomechanical modeling and urban morphology. It focuses on the development of a personalized, environment-sensitive walking model based on experimental data collected in situ.
 

Thesis objectives

The ANR CITY-STEP project aims to understand how urban morphology influences walking dynamics at the step level. The main objective of the thesis is to develop a personalized, environment-sensitive walking model capable of describing and explaining the fine-grained adaptations of pedestrian movement in real urban contexts. It is at the crossroads of biomechanics, signal processing, artificial intelligence, and urban mobility. The thesis will pursue four major scientific objectives:

• Design an experimental protocol and operate a multi-sensor device. The doctoral student will participate in the implementation and operation of an embedded measurement device that integrates inertial sensors, GNSS, a camera and eye-tracking. The challenge is to capture walking in real urban conditions without disrupting the natural behavior of the participants.
• Extracting individualized walking signatures. Using data collected in situ, the thesis will aim to detect locomotor events (steps, support phases, transitions), extract biomechanical indicators (stride variability, regularity, symmetry, micro-adjustments), and construct personalized walking signatures that incorporate interindividual variability.
• Model the adaptation of walking to the urban context. A central objective will be to analyze how environmental characteristics (slope, width, street furniture, intersections, visual stimuli) influence kinematic parameters. The thesis will develop models combining signal processing and machine learning to identify correlations between locomotor dynamics and urban morphology.
• Contributing to a multi-scale model of walking. The work will enable the proposal of a multi-scale model linking: the step scale (micro-locomotor dynamics), the urban segment scale and the street scale.

This model will be a key contribution to the CITY-STEP project and will feed into the construction of walkability indicators developed in the project.

Supervision and scientific environment

The thesis will be carried out at the GEOLOC laboratory (Gustave Eiffel University) as part of the ANR CITY-STEP project, in close collaboration with the AAU team (École Centrale de Nantes). The doctoral student will benefit from a dynamic environment combining fundamental research, experimental development, and concrete urban applications.
The GEOLOC laboratory is an internationally recognized team for its work in pedestrian navigation, inertial sensors, and smart urban mobility. It develops innovative solutions combining artificial intelligence, advanced signal processing, and embedded multi-sensor systems, applied to understanding and optimizing travel in complex environments. Involved in numerous national and European projects (ANR, H2020), GEOLOC operates within an international scientific network and collaborates with academic and industrial partners. The team offers a stimulating research environment at the crossroads of fundamental research, technological development, and societal challenges related to sustainable cities.
 

Required profile

We are looking for a candidate who is highly motivated by interdisciplinary research and interested in working at the intersection of signal processing, artificial intelligence, biomechanics, and urban mobility.

Education

Master's degree (or equivalent) in navigation/geomatics, signal processing/data science, computer science/artificial intelligence, biomechanics, or a related discipline. Excellent academic performance is expected.

Scientific and technical skills

• Solid foundation in signal processing (time/frequency analysis, filtering, event detection)
• Knowledge of machine learning (classification, supervised/unsupervised models)
• Programming skills (Python essential, C++ appreciated)
• Interest in multi-sensor systems and experimental data

Previous experience in one of the following areas would be an asset: analysis of data from inertial sensors (IMU), human activity recognition, biomechanics of walking, sensor fusion, and computer vision

Personal skills

• Interest in experimental work in real-world conditions
• Ability to work in an interdisciplinary team
• Initiative, autonomy, and scientific rigor
• Good writing skills in English
  
  

Contact
Valérie Renaudin
valerie.renaudin@univ-eiffel.fr

1- Context and scientific positioning
Independent mobility remains one of the major challenges faced by visually impaired people (VIP) in complex urban and indoor environments. While navigation technologies have significantly evolved over the last decade, most current positioning systems rely on visual sensing, external infrastructure, or satellite signals that may be unreliable or unavailable in dense urban areas and indoor spaces. Moreover, many approaches raise important concerns regarding privacy, environmental intrusiveness, or usability for visually impaired users.

Within the GEOLOC Laboratory (Université Gustave Eiffel), extensive research has been conducted on pedestrian navigation, multi-sensor positioning, and AI-based mobility analysis. Several national and European projects have contributed to the development of robust localization approaches based on inertial sensors, GNSS, and mobile sensing technologies in complex environments.

Recent advances in mobile sensing and artificial intelligence offer new opportunities for designing alternative navigation systems where sound plays a central role in spatial awareness and positioning for VIP. However, acoustic perception remains largely underexplored as a primary modality for positioning and orientation, particularly in collaborative contexts where multiple mobile devices can share acoustic information.

The proposed PhD thesis is part of the ANR MAPV project, wich aims to design an AI-driven, collaborative, sound-based positioning framework capable of improving pedestrian navigation accuracy while preserving privacy and minimizing acoustic disturbance for users and guide dog.

2- Thesis objectives
The main objective of the thesis is to develop and evaluate acoustic-based positioning methods for a collaborative positioning framework designed to support the mobility of visually impaired people. The research will lie at the intersection of signal processing, artificial intelligence, acoustic sensing, and human-centered mobility technologies.
 

• The thesis will address four main scientific objectives:
  Design minimally invasive acoustic positioning signals. Acoustic positioning techniques can provide highly accurate distance estimation, but audible signals may disturb users or interfere with the perception abilities of visually impaired people and guide dogs. The thesis will investigate acoustic signal design strategies that balance positioning accuracy, communication reliability, and minimal acoustic invasiveness.
• Develop collaborative acoustic ranging and positioning algorithms. The doctoral student will develop collaborative localization algorithms allowing multiple smartphones or wearable devices to estimate their positions using sound-based techniques. Key challenges include multi-device synchronization, signal interference, non-line-of-sight conditions, and secure information exchange between devices.
• Recognize environmental acoustic landmarks using AI. Urban environments contain many characteristic sound sources such as acoustic crossing signal, doors, elevators, electronic devices, or public infrastructure. The thesis will develop machine learning models capable of identifying and interpreting environmental sound landmarks, allowing users to improve spatial orientation, seeting acoustic landmarks and contextual awareness.
• Fuse multiple positioning sources for robust navigation. Acoustic measurements, collaborative positioning data, inertial sensors, and environmental sound recognition will be integrated within a decision-making framework capable of dynamically selecting and fusing the most reliable positioning sources. This multi-source approach aims to compensate for the drift of Pedestrian Dead Reckoning (PDR) while ensuring accurate, robust, and privacy-preserving localization.

3- Supervision and scientific environment
The thesis will be carried out at the GEOLOC laboratory (Gustave Eiffel University) as part of the ANR MAPV project. The doctoral student will benefit  rom a dynamic environment combining fundamental research, experimental development, and concrete urban applications.

The GEOLOC laboratory is an internationally recognized team for its work in pedestrian navigation, inertial sensors, and smart urban mobility. It develops innovative solutions combining artificial intelligence, advanced signal processing, and embedded multi-sensor systems, applied to understanding and optimizing travel in complex environments. Involved in numerous national and European projects (ANR, H2020), GEOLOC operates within an international scientific network and collaborates with academic and industrial partners. The team offers a stimulating research environment at the crossroads of fundamental research, technological development, and societal challenges related to sustainable cities.

4- Required profile
We are looking for a candidate who is highly motivated by interdisciplinary research and interested in working at the intersection of signal processing, artificial intelligence, acoustic sensing, and human-centered mobility technologies.

Education
Master's degree (or equivalent) in navigation/geomatics, signal processing/data science, computer science/artificial intelligence, acoustic/telecomunications, or a related discipline. Excellent academic performance is expected.

Scientific and technical skills

• Solid foundation in signal processing (time/frequency analysis, filtering, audio signal analysis)
• Knowledge of machine learning (classification, supervised/unsupervised models, deep learning)
• Programming skills (Python essential, C++ appreciated)
• Interest in multi-sensor systems and experimental data

Previous experience in one of the following areas would be an asset: analysis of data from inertial sensors (IMU), human positioning and navigation, localization/positioning methods, sensor fusion, and acoustic signal processing

Personal skills

• Interest in experimental work in real-world conditions
• Ability to work in an interdisciplinary team
• Initiative, autonomy, and scientific rigor
• Good writing skills in English

Accurate and reliable positioning is a key enabling technology for intelligent transportation systems, autonomous driving, and emerging safety and reliability-critical location-based services. While Global Navigation Satellite Systems (GNSS) combined with Real-Time Kinematic (RTK) techniques can provide centimeter-level accuracy under favorable conditions, their performance degrades significantly in dense urban environments. Signal blockage, multipath propagation, Non-Line-Of-Sight (NLOS) reception, and rapidly changing environmental conditions challenge both positioning accuracy and reliability. These are main limitations of deploying high-precision GNSS solutions in cities. One key step that enables the high precision of RTK is the GNSS carrier phase Ambiguity Resolution (AR).

The global objective of this PhD thesis is to improve the robustness, accuracy, and reliability of urban GNSS RTK navigation through learning-enhanced approaches. A focus will be given in the carrier-phase ambiguity resolution process. Nowadays, data-driven approaches are becoming increasingly popular in GNSS localization and navigation domain due to their ability to address complex challenges that traditional methods struggle to model accurately. For instance, many researchers have explored Artificial Intelligence (AI)-based techniques for detecting and excluding faulty GNSS measurements [1-2]. While these approaches have demonstrated promising results, their effectiveness in supervised learning frameworks strongly depends on the quality of labels. In practice, generating high-quality labels for urban GNSS data is often costly due to factors such as the need for human labor, expensive equipment, or access to detailed map data [3].

In this context, two closely connected research challenges will be addressed in this PhD thesis: (i) a generalizable data labeling methodology based on reinforcement learning (RL) that exploits only reference trajectories to automatically label latent reliability-related information in navigation data, such as sensor quality and adaptive measurement weights, and (ii) the integration of learned reliability information into the ambiguity resolution process to improve the performance of RTK in urban environments.

The first part of this PhD thesis therefore addresses the problem of GNSS navigation data labeling. The main research question is to develop an intelligent labeling methodology based on reinforcement learning (RL) that leverages only the reference trajectory to infer latent quantities related to GNSS measurement and ambiguity reliability. Instead of directly labeling raw observations, the proposed framework aims to automatically generate informative labels for intermediate variables that are critical to carrier-phase ambiguity resolution, such as measurement consistency indicators, ambiguity reliability states, and adaptive observation weights. By utilizing only the reference trajectory, this approach seeks to provide a cost-effective and scalable alternative to conventional labeling strategies studied in [3].

Based on the intelligent labeling framework developed in the first part of this PhD, the navigation data reliability can be labeled at different levels. The second part of this PhD topic focuses on a more specific task, which is a reliability-aware GNSS carrier phase AR assisted by Artificial Intelligence (AI). Although many research works exist in AI-based GNSS LOS/NLOS classification and pseudorange error prediction, the use of learning techniques to support carrier-phase AR remains largely unexplored. A learning architecture will be designed to first predict the reliability of the GNSS measurements. This reliability prediction will be used as an environmental perception input to support the AR process. This is the core difference with the existing literature, where the relationship between the learned features and ambiguity reliability remains largely opaque [4].

By improving high-precision positioning in dense cities, the thesis can contribute to safer automated driving, more dependable fleet operations, and accurate mapping and surveying in challenging environments. The proposed approach aims to bridge modern data-driven learning with interpretation, providing both performance gains and interpretable decision-making in safety-relevant urban scenarios.

[1] N. Zhu, R. He, Z. Wang, CarNet: A generative convolutional neural network-based line-of-sight/non-line-of-sight classifier for global navigation satellite systems by transforming multivariate time-series data into images, Engineering Applications of Artificial Intelligence, Volume 145, 2025, 110160, ISSN 0952-1976, doi.org/10.1016/j.engappai.2025.110160.

[2] García Crespillo, O., Ruiz-Sicilia, J. C., Kliman, A., & Marais, J. (2023). Robust design of a machine learning-based GNSS NLOS detector with multi-frequency features. Frontiers in Robotics and AI, 10, 1171255.

[3] Beaucamp, Benjamin, Thomas Leduc, Myriam Servières, and Ni Zhu. "Comparison of GNSS LOS/NLOS Labeling Techniques Based on a 3D model and Sky View Imagery for Soft Mobility and Vehicle Data in Urban Areas." In 2025 IEEE/ION Position, Location and Navigation Symposium (PLANS), pp. 1192-1203. IEEE, 2025.

[4] Lyu, Z., & Gao, Y. (2025). A machine learning-based partial ambiguity resolution method for precise positioning in challenging environments. Journal of Geodesy, 99, 8.
 

The research engineer will join the team of researchers at the GEOLOC laboratory and contribute to the laboratory’s research projects. In this role, he or she will be involved in continuous learning (and knowledge transfer) to address dynamic, evolving and non-stationary scenarios. He or she will also be responsible for issues relating to maps: from their real-time use to tailor approaches to mobility profiles, to the creation of specific maps and graphs dedicated to users (vulnerable or otherwise) in multimodal navigation situations.

He or she will be responsible for developing new multi-sensor positioning algorithms to overcome the weaknesses of GNSS technology in the face of radio jamming and interference.

Finally, with regard to activities supporting GEOLOC’s standardisation work, the successful candidate will be responsible for conducting and analysing signal simulations (GNSS, LEO) to help define future test scenarios for evaluating the performance of positioning systems.

Main responsibilities:

• Development of continuous and transfer learning methods to handle dynamic and non-stationary scenarios.
• Implementation of graph-based representation and reasoning approaches for navigation.
• Research into map fusion, the creation of specialised maps for specific applications (mobility for visually impaired people, multimodal mobility), and the integration of inertial maps (use of inertial landmarks).
• Integration and exploitation of data from various technologies: LEO-PNT, inertial systems, magnetometers, barometers, acoustic signals, WiFi and BLE.
• Development of multi-sensor fusion methods to improve the accuracy, robustness and resilience of localisation systems.
• Contribution to standardisation work dedicated to GNSS and emerging technologies (including LEO-PNT).
• Design and analysis of realistic test scenarios to evaluate the performance of positioning systems, including the generation and analysis of signal re-plays.

Required skills:

• Data science / AI
• Signal processing
• Navigation, Geomatics
• Fluent English
• Python, MATLAB
• Scientific supervision and writing
• Organisational skills and attention to detail
• Teamwork