Ju Wu

chinese name
Profile picture
News
Upcoming events:
Recent experiences:
  • * Software engineer intern in Merck Serono to develop semantics-driven digitalization framework for maintenance activities by integrating the digital resources and tools. 04/2022-10/2022. [Slide] [Report] [Media]
  • * R&D intern in E.M.S. Electro Medical Systems S.A. Dental Group working on multiphase simulation and continuous improvement of products. 09/2021-12/2021. [Slide] [Report] [Media]
  • * Scientific assistant in EPFL on data-driven modelling and control of mechatronic systems. 11/2022-02/2023. [Slide] [Report] [Media]
  • * Technical member of SenSwis for SensUs Challenge 2021. 08/2021-09/2021. [Slide] [Report] [Media]
About me

I am Ju Wu, a recent graduate of the MSc program in Robotics from École Polytechnique Fédérale de Lausanne (EPFL), having completed my degree in October 2022. During my time at EPFL, I had the privilege of conducting my master thesis in the ICT4M Lab under the expert guidance of Prof. Dimitrios Kyritsis and Dr. Xiaochen Zheng. Prior to this, I completed my BEng in Automation with the highest honors (5%) from Harbin Institute of Technology (HIT), where I was fortunate to be part of the Research Institute of Intelligent Control and Systems, under the mentorship of Prof. Huijun Gao. I have had the opportunity to collaborate with individuals from diverse backgrounds, which has allowed me to develop a range of skills and approaches to working with others. As a result, I am confident in my ability to contribute effectively to group projects in various capacities, whether as a team player, leader or even as a change maker. Furthermore, I have also honed my ability to work independently while balancing the interests and needs of different stakeholders. In my spare time, I enjoy hiking, swimming, and playing table tennis with friends.

Interests

  • * System modelling, simulation and control
  • * Information and communications technology for intelligent systems
  • * Signal and image processing
  • * Machine learning and optimization
  • * Biomedical applications and neuroprosthetics
  • * Social computing and biometrics

Education

  • Master of Science in Robotics. Track of Mobile Robot
    09/2020-10/2022. [Master Course Description]
    Overall 5.19/6. [Thesis] 6/6
    École polytechnique fédérale de Lausanne (EPFL)
  • Bachelor of engineering in Automation.
    09/2015-07/2019. [Bachelor Course Description]
    Overall 90.18/100 Rank 1st/37
    Harbin Institute of Technology (HIT)
Publications
  • Ju Wu, Xiaochen Zheng, Marco Madlena, Dimitris Kiritsis,"A Semantic-driven Digitalization Framework for Maintenance Strategies in the Pharmaceutical Industry,"Journal of Industrial Information Integration. Under review. [arXiv]

  • Ju Wu, Tong Wang, and Min Ma,"Finite-Time Adaptive Fuzzy Tracking Control for Nonlinear State Constrained Pure-Feedback Systems". [arXiv]

  • Ju Wu, and Tong Wang,"Adaptive Fuzzy Tracking Control for Nonlinear State Constrained Pure-Feedback Systems With Input Delay via Dynamic Surface Technique". [arXiv]

  • Tong Wang, Ju Wu, Min Ma,"Adaptive Fuzzy Tracking Control for A Class of Strict Feedback Nonlinear Systems with Time-Varying Input Delay and Full State Constraints,"IEEE Transactions on Fuzzy Systems. [Link]

  • Fan, Sijia, Tong Wang, Ju Wu, Jianbin Qiu,"Optimal containment control for a class of heterogeneous multi‐agent systems with actuator faults,"International Journal of Robust and Nonlinear Control. [Link]

Teaching
  • Control System Design and Practices. [Slide(CN)]

    The course includes Fourier Integral and Fourier Transform, Discrete Fourier Transform, Fast Fourier Transform, static error coefficient, dynamic error coefficient, noise and interference, Normal Random Variable and Normal Random Vector, Correlation Function, Spectral Density, Mean Square Error, system equivalent noise bandwidth, sensitivity and Bode integral constraints, object uncertainty, robust stability constraints, bandwidth and bandwidth design, relative stability and its indicators, servo system design, regulatory system design, disturbances observation and compensation, multi-loop system design, etc.

  • E-learning Course for Eight Pieces of Brocade Exercises. [Link]

    In my spare time, I have built a prototype of interactive training program for Eight Pieces of Brocade using Articulate 360. In the future, for the above built interactive e-learning course, there can be more interactive options, and the tactile response can be more sensitive and intuitive; AR and VR devices and techniques can be used, and sensors like tracking markers can be placed to the joints to enhance gamification, immersion and interactivity. Feel free to leave your feedback for further improvements.

Talks
  • Advancements and Trends in Graph Neural Networks: Insights from ICML 2019. [Slide(CN)]

    I delivered a presentation on the advancements and trends in Graph Neural Networks to the School of Information Engineering at Ningxia University in July 2019. The talk was based on 10 ICML articles that are closely related to GNNs.

  • Chinese Society and Islam Then and Now: Mutual Understanding, Cooperation and Common Prosperity. [Slide]

    The talk introduces the historical background and modern state of Islam in China, historical impact on culture, religion, politics and cuisine, communication and cooperation in economics, education, etc, opportunities and tips to do business in China.

Projects

master thesis
[Slide] [Report] [Media]
* Semantics-driven Digitalization Transformation Framework of Maintenance Strategies in the Pharmaceutical Industry
- Master Thesis | Python, Microsoft Power BI/Automate/App, Neo4j
  • * A diversity of the digital applications for automatic data collection, analysis, and visualization are built in need of the E&M Department of the Merck Serono to support maintenance activities on its on-site manufacturing plants in Vevey, Switzerland.
  • * Organize and integrate these semantics-driven digital tools to monitor evolution of the concerned variables, to capture the abnormal changes in the manufacturing procedures and to further support the data-driven decision making.
  • * Interact with the internal stakeholders directly and enjoy diversified business tasks such as collecting, confirming, and responding to user-requirements. Validate the developed solutions and train the business users.
proj-semester-proj-2
[Slide] [Report] [Media]
* 2D/3D Geometry Morphing via \(L_2\) Semi-discrete Optimal Transport
- Semester Project | Matlab, Python, SolidWorks
  • * Constructed the 2D/3D power diagrams constrained by the targeted meshes.
  • * Developed the semi-discrete optimal transport method based collision free geometry morphing algorithm and accelerated the codes via parallel computing.
scientific assistant 1
[Slide] [Report] [Media]
* Data-Driven Modelling and Analysis of Harmonic Drive System
- Scientific Assistant | Python, Matlab
  • * Introduce the basic mechanisms and features of harmonic drive system (HDS) and discuss the main factors that hinder the HDG transmission performance. Review literature on modeling and analysis of HDS and propose improvements for kinematic error analysis and compensation.
  • * Describe the physical laws, kinematics, dynamics, and mechanical analysis of HDS, and propose a phenomenological model based on separation of pure and flexible parts of kinematic errors. Develop and test linear and nonlinear simulation models for the flexible parts of kinematic error.
  • * Present a variety of compensation policies for HDS kinematic error and axis position tracking error and introduce a promising feedback kinematic error compensation method based on loop-shaping in the frequency domain, which is verified theoretically in preliminary configurations.
proj-intern-proj-1
[Slide] [Report] [Media]
* Multiphase Simulation for Powder Chamber
- R&D Trainee | Python, SolidWorks, MFIX, COMSOL, Cradle CFD, ANSYS Fluent
  • * Implemented multi-phase fluid simulation on the extant products and compared the sampled mass flowrate in the outlet with the simulation data to speed up design of future prototypes.
  • * The created digital twin can capture the key transient behaviors of the signals from the powder chamber and its physical testing platform.
  • * Communicated with colleagues with diverse background to get feedback and advice on improvements of current products and development of the new prototypes.
proj-competition-proj-1
[Slide] [Report] [Media]
* A Biosensor for The Rapid Quantification of Hemagglutinin-1 (H1) in Saliva
- Technical Team Member of SenSwis | Python, Linux
  • * Built a Python program that can simultaneously collect the sampled images and process them in real-time, and the total time consumed to obtain the final results is compressed to less than one third of the traditional method.
  • * Created an automated testing GUI to compute the concentration of protein with the sample of solution automatically and designed the image processing algorithms to remove noise from the captured images and count the number of specific pixels from selected regions in a list of images sampled in the given interval.
  • * Collaborated with team members to process experimental data and did curve fitting to obtain the calibration curve mapping the final intensity signals to the true concentration of the protein.
  • * Won the Analytical Performance Award and 2nd Prize of Translation Potential Award in 2021 SensUs Challenge.
proj-semester-proj-1
[Slide] [Report] [Media]
* Robotic Arm Control based on ROS via EtherCat
- Semester Project | ROS, C/C++, CMake, Linux
  • * Developed the inverse kinematics of 6-DoF robotic arm in joint space so that with the given pose of end-effector, each joint can reach the desird position.
  • * Established EtherCat communication protocol between Maxon EPOS4 micro-controller and NVIDIA Xavier Jetson and test CSP and CSV control modes with Maxon brushed motors.
  • * Generated task and joint space trajectories given the way-points and visualize the motion and trajectories of robotic arm in ROS based on Gazebo and Rviz.
proj-URP-proj-1
[Media]
* Adaptive Fuzzy Tracking Control for a Class of Strict-Feedback Nonlinear Systems With Time-Varying Input Delay and Full State Constraints
- Undergraduate Research Assistant | Matlab, C/C++
  • * Designed an auxiliary integrator for adaptive fuzzy controller to compensate the negative influence imposed by stochastic input delay.
  • * Proposed a novel finite-time adaptive fuzzy tracking control strategy for uncertain pure-feedback systems with full state constraints. Carried out feasibility check to find out feasible parameters of controllers through solving a static semi-infinite nonlinear constrained problem numerically.
  • * Collaborated with colleagues to compile part of research results and published work on IEEE T ransactions on Fuzzy Systems DOI: 10.1109TFUZZ.2019.2952832.
proj-course-ml4f
[Report]
* Calibration of Local (Stochastic) Volatility Model via Neural SDEs
- Machine Learning for Finance | Course Project | Python, PyTorch, Jupyter Notebook
  • * The classical finance modelling procedure involves handcrafting a fixed parametric model, followed by calibration and verification, which neglects the issues of model selection and uncertainty. In this project, an over-parameterized neural network is used to represent the drift and diffusion components of the model, allowing simultaneous calibration and data-driven model selection.
  • * The project aims to fit local volatility and local stochastic models to option price data with high accuracy of calibration using minimum MSE, and uses the surface stochastic volatility inspired (SSVI) model as a benchmark to fit the implied volatility data. Feedforward neural networks are also used to over-parameterize the models and control the hedging strategy.
proj-course-smn
[Slide] [Report]
* Decode wrist and finger movements using a new Medium Density EMG prototype
- Sensorimotor Neuroprosthetics | Course Project | Python, PyTorch, Jupyter Notebook
  • * The work aims to decode wrist and finger movements using electromyography and machine learning techniques.
  • * Challenges include selecting the best technique, creating a universal classification model, and calibrating for electrode placement changes.
  • * Results show high accuracy in classifying movements for multiple patients, and transfer learning and calibration can improve classification under the new patients and electrode shifting respectively. Further study is needed with a larger number of patients and electrode orientations.
proj-course-rte
[Report]
* Neural Network Acceleration for Image Recognition of Digits
- Real-time Embedded Systems | Course Project | VHDL, C
  • * Developed a working multi-master system, capable of classifying handwritten digits using a hardware neural network accelerator on FPGA. The accelerator is implemented in VHDL and exploits burst capabilities to read the image contents, and parallelized operations for inference.
  • * The neural network inference is sped up by a factor of 75 \(\times\) with the hardware accelerator compared to software.
proj-course-nmb
[Slide] [Report]
* Numerical Model of a Tubular Bioreactor for Stem Cells
- Numerical Methods in Biomechanics | Course Project | COMSOL, Matlab
  • * Built numerical model of a tubular bioreactor for the growth and differentiation of human mesenchymal stem cells, verify and validate it with data from previous papers.
  • * Set material properties and physical laws it conforms to, use FEM to generate computational model and do mesh convergence study.
  • * Do simulation and obtain the appropriate configuration of the alginate beads supporting cell growth to enable homogeneous oxygen diffusion, and report physiologically relevant shear stresses at the beads surface.
proj-course-aml
[Report]
* Reliability Evaluation of Reinforcement Learning Algorithms
- Advanced Machine Learning | Course Project | Stable-Baselines3, Python, Jupyter Notebook
  • * Do experiments on A2C and PPO2 for lunar lander and cartpole tasks based on GoogleAI baselines with customized hyper-parameters and settings.
  • * Evaluate the reliability of the RL algorithms with dispersion&risk across runs&time metrics.
proj-course-rp
[Report1] [Report2] [Report3]
* ABB IRB 120 Robotic Arm Control; Motion Capture Systems; 2-DOF AERO Helicopter Control
- Robotics Practicals | Course Project | Matlab, Simulink, ABB RAPID
  • * The motion capture system practical explored optical motion capture as a powerful tool for recording and analyzing body motion, and highlighted its value in state estimation of robots, but also emphasized the importance of careful calibration and marker positioning due to its limitations.
  • * The helicopter control practical aimed to identify and control a 2 DOF AERO setup using LQR and MPC controllers, apply previously learned control and identification techniques, modify plant parameters, test controller performance, and explore methods to enhance plant performance.
  • * The ABB robotic arm control practical involves an introduction to the ABB IRB 120 industrial robot, including theoretical details on its properties and limitations, followed by a practical component where participants operated the robot and conducted two experiments to explore different control manners, including manually programming the robot to follow a line and implementing a 2 player TicTacToe game using the RAPID language through the PC interface.
proj-course-dl
[Report1] [Report2]
* Creating Neural Network from Scratch With Basic Tensor Operations
- Deep Learning | Course Project | Python, PyTorch
  • * Created a neural network framework for deep learning from scratch comprised basic modules such as dense layers, activation functions (Relu, Tanh and Sigmoid), loss functions (MSE, BCE and CE) and optimizers (SGD and Adam).
  • * Build a network with the designed framework to do binary classification task and use He initialization to solve the vanishing gradients and local minimum induced by increased depth and width.
proj-course-iapr
[Report] [Slide]
* Cascade Feature Extraction Method for Digital Numbers
- Image Analysis and Pattern Recognition | Course Project | Python, Scikit-learn, Jupyter Notebook
  • * Created a cascade feature extraction method in virtue of Fourier descriptors and PCA method to reduce and embed the high dimensional features.
  • * Tested the proposed method on digital numbers from MNIST database and achieve similar performance than features manually selected in Fourier descriptors but the cascade method automatize the manual selection procedure for feature.
  • * Prepared to generalize the method via exploring the more underlying mechanism of automatic optimal feature extraction and employ it in other applications.
proj-course-mpc
[Report]
* Quadrotor drone MPC Algorithm
- Model Predictive Control | Course Project | Matlab, Gurobi
  • * Developed linear and nonlinear MPC tracking control algorithm for quadrotor and designed the test benchmark environment to compare performance of them under different conditions and visualize the obtained results.
  • * Used GUROBI solver in YALMIP to solve quadratic programming problems constructed when deigned the optimal controllers with finite horizon.
  • * Designed the state observer for linearized mutually independent subsystems with unknown external disturbance.
proj-course-lr
[Report] [Slide]
* Reinforcement Learning Control of Biped Robot
- Legged Robots | Course Project | Matlab
  • * Developed RL training and simulation environment for biped robot agent based on TD3 networks in Matlab toolbox.
  • * Design the customized reward function to achieve high speed and low cost of transport in the long run.
  • * Achieve better performance than other state-of-art methods such as cyclic pattern generators, MPC and inverse dynamics.
proj-course-acs
[Report1] [Report2] [Report3]
* Robust Control of an Active Suspension System
- Advanced Control Systems | Course Project | Matlab, Simulink, YALMIP
  • * Designed a data-driven mixed sensitivity robust controller with required specification for active suspension system with multiplicative uncertainty.
  • * Used YALMIP to solve the linear and semi-infinite programming problems in the optimization of controller that guarantees internal stability.
  • * Wrote basic pole placement function for RTS controller design and implemented the switching adaptive control for active suspension system.
proj-course-si
[Report1] [Report2] [Exam]
* Nonparametric and Parametric Identification Methods
- System Identification | Course Project | Matlab
  • * Nonparametric methods include fourier analysis and spectral analysis.
  • * Parametric methods include FIR, ARX, OE, ARMAX, Box-Jenkins polynomial model (BJ), subspace method to estimate state-space model, four-stage instrumental variable method (IV4), etc.
  • * Conducted system identification procedure from input data analysis, frequency response analysis, order estimation for the system and the delay, model selection based on fitting results in time and frequency domain, validation through statistical tests consisting of the uncorrelation test and a whitness test.
proj-course-odm
[Report]
* Robust Regression for Life Expectancy Data from WHO
- Optimal Decision Making | Course Project | Matlab, Python
  • * Constructed the best estimator for the worst-case training set for adversarial training.
  • * Converted the MinMax problem for robust regression to linear programs, using convex hull and duality.