M2 Mechatronics, Machine Vision and Artificial Intelligence

1. Introduction to Reinforcement Learning (RL)

Definition and motivation

RL framework: agent, environment, states, actions, rewards

Markov Decision Processes (MDP) and Bellman equations

Exploration vs. exploitation

2. Classical RL Methods

Dynamic Programming: policy evaluation, policy iteration, value iteration

Monte Carlo methods

Temporal Difference learning: SARSA, Q-Learning

3. Policy Gradient and Actor–Critic Methods

Value-based vs. policy-based methods

REINFORCE algorithm and stochastic policy gradients

Actor–Critic architecture and advantage estimation (A2C, A3C)

4. Deep Reinforcement Learning (Deep RL)

Function approximation using neural networks

Deep Q-Network (DQN) and its extensions (Double DQN, Dueling DQN, PER)

Deep policy optimization: PPO, TRPO, DDPG, TD3, SAC

Training stability, reward shaping, and exploration strategies

5. Model-Based and Hierarchical RL

Model-based planning: Dyna-Q, world models

Hierarchical RL: options framework, sub-policy learning

Meta-learning in RL

6. Multi-Agent Reinforcement Learning

Cooperative and competitive settings

Communication and coordination among agents

Applications to swarm robotics and game environments

7. Applications of Deep RL

Robotics control and motion planning

Autonomous driving and drone navigation

Game playing (Atari, Go, StarCraft)

Decision making in finance and resource optimization

8. Recent Trends and Advanced Topics

Offline RL and imitation learning

Safe and explainable RL

Curriculum learning and lifelong RL

Foundation models for control (e.g., Gato, RT-2, Diffusion Policies)

1. Introduction to 3D Data

Nature and sources of 3D data: depth sensors, LiDAR, stereo vision, multi-view reconstruction

3D data representations: point clouds, meshes, voxel grids, implicit functions

Coordinate systems and geometric transformations

2. 3D Data Acquisition & Preprocessing

Camera models and calibration

Depth estimation and stereo reconstruction

Point cloud filtering, registration, and alignment (ICP, RANSAC)

Normal estimation and surface reconstruction

3. 3D Feature Extraction & Classical Methods

Geometric descriptors: curvature, normals, shape histograms (FPFH, SHOT)

Keypoint detection in 3D

Object segmentation and classification with traditional approaches

4. Deep Learning for 3D Data

3D convolutional networks (3D CNNs, volumetric approaches)

Point-based architectures: PointNet, PointNet++, DGCNN

Graph-based methods for 3D learning

Mesh and implicit representation learning

5. Multi-View and Hybrid Approaches

Multi-view CNNs and differentiable rendering

Neural Radiance Fields (NeRF) and implicit scene representation

2D–3D joint learning and fusion techniques

6. 3D Scene Understanding

Semantic and instance segmentation of 3D scenes

Object detection and pose estimation in 3D

• Strong background in mechatronics, robotics, control, and embedded systems.
• Basic understanding of machine learning and AI concepts.
• Proficiency in MATLAB, Python, or C++.
• Experience with modeling, simulation, or hardware prototyping.
• Ability to conduct literature review and scientific writing.

•  Introduction to research methodology: scientific process, hypotheses, validation.
• Literature review techniques: databases, keyword strategies, scientific sources.
• Project definition: problem formulation, objectives, expected outcomes.
• Experimental design: modeling, simulation, prototyping, data collection.
• Integration of mechatronics and AI: sensing, perception, control, learning.
• Data analysis, metrics, and evaluation of results.
• Research ethics, reproducibility, and documentation.
• Weekly supervision meetings and progress monitoring.

• Develop the ability to design, plan, and conduct an independent research project.
• Apply advanced mechatronic and AI techniques to solve a real scientific or industrial problem.
• Strengthen skills in experimental design, data analysis, and model validation.
• Enhance scientific writing and presentation skills.
• Learn to manage a research schedule, milestones, and deliverables.
• Produce a high‑quality research report equivalent to a pre‑publication draft.

• Autonomous research planning and execution.
• Advanced problem‑solving in mechatronics and AI.
• Scientific writing and literature review competencies.
• Experimental design, data acquisition, and quantitative analysis.
• Mastery of simulation and/or prototyping tools.
• Communication, project management, and research ethics.

1. Introduction to Machine Learning

Overview of machine learning paradigms: supervised vs. unsupervised

Key concepts: overfitting, generalization, bias–variance tradeoff

2. Data Preparation & Representation

Data preprocessing: normalization, feature scaling, encoding categorical variables

Feature engineering and dimensionality reduction (PCA overview)

Train/validation/test split and cross-validation

3. Linear Models

Linear regression, logistic regression

Regularization: Ridge, Lasso, Elastic Net

Model evaluation metrics (classification & regression)

4. Kernel Methods

Concept of feature space and kernel trick

Common kernels (linear, polynomial, RBF, sigmoid)

Kernelized algorithms: Support Vector Machines, Kernel Ridge Regression

5. Tree-Based Methods

Decision Trees: impurity measures, pruning, interpretability

Ensemble tree methods: Random Forests, Gradient Boosted Trees, XGBoost

Feature importance and model explainability

6. Ensemble and Hybrid Methods

Bagging, boosting, stacking

Comparison between linear, kernel, and tree-based methods

7. Neural Networks (Intro)

Perceptron and multilayer perceptron (MLP)

Backpropagation and gradient-based optimization

Comparison with kernel and tree-based methods

8. Model Selection & Evaluation

Cross-validation and hyperparameter tuning

Regularization strategies and early stopping

Bias–variance analysis and learning curves

9. Practical Sessions

Implementation with Python/scikit-learn

Model training, tuning, and evaluation on benchmark datasets

Mini-project: compare linear, kernel, and tree-based methods

1. Overview of Advanced ML

Recap of supervised learning and neural network fundamentals

Motivation for deep learning: hierarchical feature learning

GPU computing and modern frameworks (TensorFlow, PyTorch)

2. Deep Feedforward Networks

Architecture and training of deep networks

Activation functions, normalization layers, dropout

Optimization: gradient descent variants, learning rate scheduling

3. Convolutional Neural Networks (CNNs)

Convolutional layers, pooling, and feature maps

Classical architectures: LeNet, AlexNet, VGG, ResNet

Applications in vision: classification, detection, segmentation

4. Recurrent and Sequence Models

Recurrent Neural Networks (RNN), LSTM, GRU

Attention mechanisms and sequence-to-sequence models

Applications: speech recognition, time-series forecasting, NLP

5. Transformer Architecture and Large Language Models (LLMs)

The Transformer: self-attention, multi-head attention, positional encoding

Pre-training and fine-tuning paradigms

Overview of major LLMs (BERT, GPT family, T5)

Prompt engineering and transfer learning concepts

6. Generative Deep Learning

Autoencoders and Variational Autoencoders (VAEs)

Generative Adversarial Networks (GANs)

Diffusion models (intro and applications)

7. Modern Techniques & Trends

Self-supervised and contrastive learning

Foundation models and multi-modal learning

Efficient training: pruning, quantization, model compression

8. Evaluation & Ethics

Evaluation of deep learning models (accuracy, robustness, interpretability)

Bias, fairness, and ethical considerations in AI

Responsible deployment and environmental cost of large models

• Basic mechatronic engineering knowledge.
• Multi‑physics modeling (mechanical, electrical, control).
• Embedded systems knowledge and C/Python programming.
• Control engineering: PID, digital control.
• Proficiency with MATLAB/Simulink and introduction to Simscape/Modelica.

•  Introduction to advanced mechatronic systems: trends and industrial applications.
• Advanced actuators: torque motors, smart actuators,high‑performance hydraulics.
• Advanced sensors: vision systems, IMUs, force/torque sensing, sensor fusion.
• multi‑physics modeling: electromechanical, hydraulic, and thermal domains.
• High‑performance embedded systems: STM32, DSP, FPGA (introduction).
• Software architectures: ROS/ROS2, distributed control, real‑time communication.
• Advanced control: optimal control, robust control (intro), dynamic feedforward.
• Digital twins and Model‑Based Design: concept, modeling, validation.
• Intelligent mechatronic systems: perception, decision, and action integration.
• Advanced simulation: Simscape, Modelica, real‑time solvers.
• Case studies: exoskeletons, advanced robotic systems, mobile platforms.
• Final project: design of an advanced mechatronic subsystem.

• Understand advanced technologies used in modern mechatronic systems.
• Model and analyze complex multi‑domain physical systems.
• Study and integrate advanced actuators and sensors.
• Design high‑performance hardware and software architectures.
• Simulate and optimize multi‑physics mechatronic systems.
• Develop a mechatronic prototype integrating perception, control, and actuation.
• Use modern tools such as Simscape, ROS, embedded boards, and industrial drivers.

• Design and modeling of advanced mechatronic systems.
• Mastery of multi‑physics tools for simulation and analysis.
• Integration of advanced actuators/sensors and real‑time communication.
• Development on high‑performance embedded architectures.
• Design of complex hardware and software architectures.
• Skills in optimization and performance analysis.
• Teamwork, project management, and technical communication.

1)Mechatronics with Experiments by Cetinkunt & Sabri University of Illinois & Chicago

2)Siciliano & Khatib – Springer Handbook of Robotics.

**3)**Mechatronics: Electronic Control Systems in Mechanical and Electrical Engineering Paperback – March 15, 2019

by William Bolton (Author)

• Mathématiques avancées : algèbre linéaire, matrices, dérivées partielles.
• Mécanique analytique : notions de base en mécanique du solide.
• Automatique et systèmes dynamiques : modélisation, équations d'état.
• Programmation scientifique : MATLAB ou Python.
• Notions de robotique : repères, transformations homogènes

• Introduction : rappels de mécanique du solide, espaces de configuration.
• Cinématique géométrique : transformations homogènes, rotations.
• Cinématique directe et inverse : chaînes ouvertes, singularités.
• Cinématique différentielle : vitesses, accélérations, jacobien.
• Dynamique Newton–Euler : formulation récursive.
• Dynamique Lagrangienne : équations de Lagrange, matrices d'inertie.
• Prise en compte des effets réels : frottements, flexibilités, efforts externes.
• Dynamique en espace opérationnel : matrices d’inertie opérationnelle.
• Simulation numérique : solveurs MATLAB/Simscape/Python.
• Applications : manipulateurs, robots mobiles, exosquelettes.

• Maîtriser la cinématique directe et inverse des systèmes mécatroniques et robotiques.
• Établir les équations de la cinématique différentielle via les matrices Jacobiennes.
• Développer les modèles dynamiques basés sur Newton–Euler et Lagrange.
• Intégrer les frottements, flexibilités et perturbations dans les modèles.
• Analyser et simuler le comportement dynamique de systèmes mécatroniques.
• Appliquer les modèles dynamiques pour la commande avancée des robots.
• Utiliser MATLAB/Python pour simuler des modèles cinématiques et dynamiques.

• Modélisation cinématique de systèmes robotiques complexes.
• Analyse des vitesses/accélérations via les Jacobiennes.
• Formulation dynamique complète (Newton–Euler, Lagrange).
• Prise en compte des phénomènes réels (frottements, inerties, perturbations).
• Simulation numérique avancée.
• Conception de modèles utiles à la commande avancée.
• Maîtrise de MATLAB/Simulink et Python.
• Communication scientifique et rédaction technique.

- Spong, Hutchinson & Vidyasagar – Robot Modeling and Control.

-Siciliano & Khatib – Springer Handbook of Robotics.

Circuit theory, basic analog & digital electronics

1. Measurement & Instrumentation: Architecture of measurement chains; performance metrics.
2. Transduction: Physical-to-electrical conversion; overview of ADC/DAC.
3. Errors & Noise: Systematic/random errors; noise sources; uncertainty and noise reduction.
4. Differential Measurements: Instrumentation amplifiers; CMRR; common-mode effects.
5. Sensors: Main classes of sensors and conditioning circuits.
6. Modulation Techniques: Chopper stabilization, auto-zero, AM/FM, improving low-frequency performance.
7. Coherent Detection: Lock-in amplifier principles and applications.
8. Measurement System Improvement: Optimization of accuracy, bandwidth, and robustness.

Labs & Tutorials (Summary)

• Sensor characterization
• Instrumentation amplifier design and Filters
• Noise and CMRR measurement
• Digital lock-in implementation
• Design of a simple measurement system

This module introduces students to the fundamentals of modern measurement systems and electronic instrumentation. It focuses on how physical quantities are converted, conditioned, and processed, and on how to evaluate and improve the accuracy, noise performance, and stability of electronic measurement chains. At the end of the course, students will understand key sensors, amplification techniques, error and noise mechanisms, and advanced methods such as modulation and coherent detection, that enhance low-frequency measurements.

1. D. A. Johns & K. Martin , Analog Integrated Circuit Design, Wiley.
2. R. Pallas-Areny & J. G. Webster, Sensors and Signal Conditioning, Wiley.
3. J. Fraden , Handbook of Modern Sensors, Springer.
4. H. K. P. Neubert , Instrument Transducers: An Introduction to Their Performance and Design, Oxford.

• Basic robotics: kinematics, coordinate frames, transformations.
• Programming in Python and/or C++.
• Understanding of embedded systems fundamentals.
• Knowledge of control systems and real-time constraints.
• Familiarity with Linux and command-line tools is recommended.

•  Introduction to robotic system architectures: layered and modular design.
• Middleware concepts: nodes, topics, services, actions, DDS.
• ROS/ROS2 fundamentals: packages, launch files, parameter server, TF.
• Sensor and actuator integration: IMUs, cameras, LiDAR, motor drivers.
• Communication and networking: real-time constraints, QoS, distributed systems.
• Robotic software pipelines: perception, mapping, planning, control.
• Embedded robotics: microcontrollers, SBCs (Raspberry Pi, Jetson), deployment.
• Simulation environments: Gazebo, RViz, ROS2 simulation workflows.
• Safety, debugging, testing, and performance monitoring.
• Case studies: mobile robots, manipulators, autonomous platforms.
• Project: building a full robotic architecture in ROS/ROS2.

• Understand the structure and design of modern robotic system architectures.
• Master middleware concepts and communication frameworks such as ROS/ROS2.
• Develop modular robotic software pipelines (perception–planning–control).
• Implement and deploy robotic programs on embedded and distributed systems.
• Handle real-time constraints, inter-process communication, and system integration.
• Design and prototype a complete robotic software stack.
• Gain hands-on experience with ROS/ROS2, simulation tools, and hardware interfaces.

• Design and implementation of modular robotic architectures.
• Programming and integration using ROS/ROS2.
• Real-time communication and distributed system design.
• Perception‑planning‑control pipeline development.
• Embedded deployment and hardware‑software integration.
• Debugging, optimization, and performance evaluation.

1. Quigley et al. – Programming Robots with ROS.
2. S. Chitta, E. Marder-Eppstein, W. Meeussen, V. Pradeep, A. R. Tsouroukdissian, J. Bohren, D. Coleman, B. Magyar, G. Raiola, M. Lüdtke, E. F. Perdomo, ros_control: A generic and simple control framework for ROS. J. Open Source Soft. 2, 456 (2017).
3.  
   • Intel & Open Robotics documentation for ROS2 DDS/QoS.

Linear Control.

Mots clés
optimal control, Lyapunov, Linear control, back-stepping, Robust control
Description du contenu de l'enseignement
Abstract
This course covers nonlinear control design and analysis methods used on Unmanned Aerial Systems. It covers robust stability analysis, optimal control and non-linear control theory used to project state feedback into output feedback. This part forms baseline control algorithms that are augmented with adaptive control in the second part. Lyapunov stability theory is followed by an introduction to the design and analysis of adaptive control systems. Key design points are discussed and illustrated through simulation examples. This course ends with an overview of open problems and future research directions.

Course outline
- Flight modes and Linear control
- Robust control
- Optimal control
- Lyapunov approach
- Back-stepping
- Adaptive control
- Open problems.

W. Durham 'aircraft flight dynamics and control', Wiley 2013 B. Etkin and L.D. Reid. Dynamics of flight. Stability and control. Wiley,1996. D. P. Raymer. Aircraft design : a conceptual approach. AIAA, 1989. G. F. Franklin et al ‘ Feedback control of dynamic systems ’ AddisonWesley,1991 D. McLean ‘ Automatic flight control systems ’ Prentice Hall, 1990 J. F. Magni et al ‘ Robust flight control ’ Springer, 1997.

• Solid mechanics: kinematics and dynamics of rigid bodies.
• Advanced mathematics: linear algebra, Jacobians, nonlinear equations.
• Robotics fundamentals: coordinate frames, transformations, direct/inverse kinematics.
• Control and dynamic systems: Lagrange and Newton–Euler basics.
• Programming skills in MATLAB and/or Python.

• Introduction to parallel manipulators: advantages, architecture families, applications.
• Mobility analysis: constraints, degrees of freedom, criteria.
• Direct kinematics: nonlinear systems, existence and uniqueness issues.
• Inverse kinematics: analytical and numerical methods, workspace analysis.
• Differential kinematics: Jacobian matrices, velocity relations, singularity types.
• Kinetostatic modeling: stiffness, compliance, and force transmission.
• Dynamic modeling using Lagrange formulation for closed-loop mechanisms.
• Newton–Euler formulation adapted to parallel structures.
• Handling constraints: loop-closure equations, holonomic constraints.
• Workspace and singularity analysis
• Numerical simulation: ODE solvers, simulation of trajectories, dynamic responses.
•  Case studies: Stewart platform, Delta robot, 6-DoF parallel manipulators.
• Mini-project: modeling and simulation of a complete parallel robot.

• Understand the structure, characteristics, and classifications of parallel robots.
• Master direct and inverse kinematics of parallel manipulators.
• Formulate Jacobians and analyze singularities specific to parallel architectures.
• Develop dynamic models using Lagrange and Newton–Euler methods.
• Model constraints, closed-loop kinematics, and mobility of parallel mechanisms.
• Simulate parallel manipulators using MATLAB/Python and numerical solvers.
• Apply modeling results to control, design optimization, and performance evaluation.

• Ability to model kinematic and dynamic behavior of parallel robots.
• Mastery of closed-loop kinematic constraints and their implications.
• Use of numerical tools for simulation and validation.
• Understanding singularity analysis and workspace characterization.
• Capacity to apply models for control and design optimization.

-Parallel Robots, Merlet, J.P., Livres

- Tsai, Robot Analysis: The Mechanics of Serial and Parallel Manipulators.

- Angeles, Fundamentals of Robotic Mechanical Systems.