Ph.D. · Robotics & Intelligent Perception

Ayman
Beghdadi

Robotics · Computer Vision · Artificial Intelligence
Ph.D. from Université Paris-Saclay · Academic research and industrial experience in robotic perception and immersive systems
Dynamic Visual SLAM · Scene Understanding · Graph Neural Networks · Multimodal Perception · Human–Robot Interaction · Embedded AI
AB
Ayman Beghdadi
3 + 6
Journal + conference publications
190.5 h
Teaching (Eq. TD)
5
Projects & internships supervised
27 · 61
French CNU qualifications

Robust perception for autonomous & interactive systems

Robotics · Vision · Embedded Systems

Research spanning academic prototypes, real robotic platforms and industrial immersive systems.

I am a researcher in robotics, computer vision and artificial intelligence. I received my Ph.D. in robotics, computer vision and AI from Université Paris-Saclay, where my doctoral work focused on multisensory interaction between multidimensional, multi-sensor robotic systems and users.

My research focuses on perception methods that remain robust in complex, dynamic and uncontrolled environments. My work combines geometric computer vision, deep learning, semantic scene understanding, graph-based representations and multimodal sensing, with a strong emphasis on real-time deployment.

Alongside academic research, I have industrial experience at Thales on virtual and mixed-reality systems, human factors, cybersickness, multimodal user-state acquisition and computer-vision solutions for robust immersive systems.

Scientific focus

My work sits at the intersection of robotic perception, computer vision and embedded intelligence, with the goal of making autonomous or interactive systems more robust, context-aware and deployable.

01

Dynamic Visual SLAM

Geometric and semantic localization and mapping in dynamic indoor/outdoor environments, including depth-aware reasoning and dynamic feature rejection.

02

Scene Understanding & GNNs

Lightweight graph-based scene representations combining semantic and spatial object information for contextual classification and adaptive perception.

03

Robust Computer Vision

Benchmarking and improving object detection under realistic distortions, including context-aware photorealistic image and video degradations.

04

Multimodal Perception

Fusion of visual, inertial, physiological and gaze-related signals to infer user state and support context-aware interaction.

05

Human–Robot Interaction

Multisensory interaction, mixed reality, robot platforms and user-centered perception for interactive and assistive systems.

06

Embedded & Real-Time AI

Integration of perception pipelines on real systems with attention to latency, resource constraints, sensors, testing and software robustness.

Academic and industrial path

2023 — present
Virtual Reality Solution Architect
Thales · Cergy, France
R&D on human factors and cybersickness in flight simulation, multimodal context analysis, robust computer vision for mixed-reality systems, and acquisition pipelines combining IMU, physiological measurements and gaze estimation.
2023 — 2026
Post-Ph.D. research activity
Université Paris-Saclay · IBISC and international collaborations
Dynamic visual SLAM, scene classification with graph neural networks, multimodal behavior analysis and scientific software/dataset dissemination.
2020 — 2023
Ph.D. in Robotics, Computer Vision & Artificial Intelligence
Université Paris-Saclay · IBISC (EA 4526)
Thesis on multisensory interactions between a multidimensional, multi-sensor robotic system. Doctoral work completed in 34 months together with university teaching activities.
2018 — 2020
Master E3A — Mobile Automatic Systems
Université Paris-Saclay
Graduated with honors and ranked first in the program. Final project on multisensory interaction in mixed reality with a multidimensional robotic platform.
2017
Bachelor's degree — Electronics, Electrical Energy & Automation
Université Paris-Sorbonne · Robotics option

Selected projects

Repositories and research projects connected to robust visual perception, scene understanding and reproducible evaluation.

G

Hybrid-GCNN-CNN

PyTorch Geometric implementation of a lightweight hybrid graph/CNN scene-classification framework using detection outputs to build semantic-spatial scene graphs.

Python · PyTorch Geometric · YOLACT · Scene Graphs
View repository →
D

CD-COCO

Context-aware extension of MS-COCO with local and global photorealistic distortions for robustness evaluation and training of vision models.

MATLAB · Dataset · Robust Computer Vision
View repository →
B

Distorted Natural COCO

Benchmark subsets for object detection under naturally occurring blur, noise, rain, backlight, contrast changes and other real degradations.

Python · COCO Evaluation · Robustness
View repository →
S

FAE-SLAM

Ongoing dynamic RGB-D visual SLAM research combining geometry, semantics, depth-aware evidence, scene context and real-time constraints.

C++ · OpenCV · RGB-D SLAM · Deep Learning
V

PVD4RCV

Photo-realistic multi-distortion video dataset and evaluation framework for benchmarking robust computer-vision models.

Video Dataset · Robust Vision · VCIP 2025
Paper / DOI →
C

CD-COCO collaborative repository

Collaborative repository associated with CD-COCO and scene-context-aware robustness experiments on the aymanbegh account.

MATLAB · Collaboration
View repository →

Research output

Selected peer-reviewed work in dynamic visual SLAM, robust computer vision, scene understanding and multimodal robotics.

2025
PVD4RCV: A Photo-Realistic Multi-Distortion Video Dataset for Benchmarking and Developing Robust Computer Vision Models
A. Beghdadi, M. Ullah, et al.
IEEE VCIP 2025
DOI →
2024
A New Lightweight Hybrid Graph Convolutional Neural Network–CNN Scheme for Scene Classification Using Object Detection Inference
A. Beghdadi, A. Beghdadi, M. Ullah, F. Alaya Cheikh, M. Mallem
IEEE EUVIP 2024
DOI →
2023
CD-COCO: A Versatile Complex Distorted COCO Database for Scene-Context-Aware Computer Vision
A. Beghdadi, A. Beghdadi, M. Mallem, L. Beji, F. Alaya Cheikh
IEEE EUVIP 2023
DOI →
2023
DAM-SLAM: Depth Attention Module in a Semantic Visual SLAM Based on Objects Interaction for Dynamic Environments
A. Beghdadi, M. Mallem, L. Beji
Applied Intelligence 53(21), 25802–25815
DOI →
2022
A Comprehensive Overview of Dynamic Visual SLAM and Deep Learning: Concepts, Methods and Challenges
A. Beghdadi, M. Mallem
Machine Vision and Applications 33(4), 54
DOI →

From fundamentals to research projects

Teaching experience from undergraduate to Master's level, covering programming, embedded systems, signal processing, numerical methods, robotics and computer vision.

P

Programming

Java, Python, C/C++, algorithmic reasoning and software development.

E

Embedded & digital systems

VHDL, real-time Unix, hardware/software integration, sensors and embedded architectures.

R

Robotics & perception

Computer vision, robot perception, simulation, mixed reality and student research supervision.

Methods & tools

Programming & AI

C/C++ Python Java MATLAB PyTorch TensorFlow Graph Neural Networks Deep Learning

Robotics & Vision

ROS OpenCV Visual SLAM RGB-D Object Detection Segmentation Scene Graphs Sensor Fusion

Systems & integration

Linux VHDL CANopen Embedded Systems Real-Time Systems Multisensor Acquisition

Immersive & human-centered systems

Virtual Reality Mixed Reality Gaze Estimation Human Factors Cybersickness Multimodal Analysis

Research & collaborations

For research collaborations, academic projects and discussions in robotics or computer vision, you can contact me through the channels below.