About me
I am a computer scientist with a robotics background.
In 2019 I became a master student member EIT Digital, studying Autonomous systems - Robotics and Machine learning. I did the first master year (2019-2020) at Technical Universit of Berlin, with 3.35 / 4.00 GPA. On July 2021 I finishied my master at Aalto University, School of Electrical Engineering, in Autonomous systems Robotics & AI. Since January 2021 I have been working as a researcher at the Finnish Geospatial Research Institute (FGI) under the National Land Survey of Finland (NLS).
In August 2018 - September 2019 I worked at Multisoft Constanta as Full-stack software engineer. There I developed skills in analysis, design and implementation of web and desktop applications using C#, .NET, SQL Server, Javascript (Bootstrap & jQuery), provide cloud solutions for hotels and restaurants.
In 2017-2019 I obtained a master's degree in "Distributed Multi-Model Virtual Environments" at Ovidius University of Constanta with final thesis grade: 9.66 / 10 and overall average grade: 9.78 / 10 In 2015-2018 I did my bachelor in Mathematics and Computer Science, at Ovidius University of Constanta. I graduated in Mathematics and Computer science with final thesis grade: 10 / 10 and overall average grade: 9.87 / 10. In 2014 I did my first bachelor's degree in economics and business administration, at Ovidius University of Constanta. I graduated in 2017 with final bachelor thesis grade: 10/10 and overall average grade: 8.57 / 10.
Awards and Honors
-
1st prize ESTIC 2018 - Speech and Speaker recognition app for Romanian language, EDITION XI, Contest for Students in Information Technology.Link
-
1st prize ESTIC 2019 - Mobile robot embedded with object detection and recognition app, EDITION XII, Contest for Students in Information Technology, Link
-
UNIVERSITTY OVIDIUS OF CONSTANTA, SUMMER SCHOOL, July 2017 | Constanta, Romania Cerva Summer school on virtual environments, cultural heritage 6’th edition. Link
-
KTH ROYAL INSTITUTE OF TECHNOLOGY, SUMMER SCHOOL August 2020 | Stockholm, Sweden BIG DATA ANALYTICS - EIT Digital summer school, Link
THESIS & PUBLICATIONS
-
Speech & Speaker recognition for Romanian Language, Published in RoCHI 2018, Computer Science, Romania.
Recommendations
-
PhD. Associate Professor of Computer Science DRAGOS SBURLAN, Constanta, 08/01/2019
-
Conf.univ.dr. ELENA PELICAN, Constanta, 25/11/2018
Master following topics
- Object-Oriented Programming, Dynamic programming
- Relational databases, Front-end, Back-end development
- Machine learning
- Unsupervised
- PCA, Outlier Detection, LLE
- Clustering, K-means, GMM, Expectation-Maximization, Hierarchical clustering
- Supervised
- Kernel Ridge Regression, Cross-Validation, Logistic regession
- Support vector machines, Boosting
- Neural networks, CNN, LSTM
- Feature learning, selection
- Preference learning, ranking
- Reinforcement learning
- Value Iteration, Policy Iteration
- Q-Learning, DQN
- Policy Gradient, Actor critic
- Unsupervised
- Computer vision
- Image formation and processing
- Feature detection and matching
- Structure from motion, stereo and 3D reconstruction
- Object recognition and detection
- Robotics
- Forward and Inverse Kinematics
- Path planning & tracking
- Filtering, localization, tracking, mapping
- Kalman Filter, EKF, UKF, Particle filter
- ROS, SLAM, Navigation
- Image-based, Position-based visual servoing
SKILLS
- PROGRAMMING
- Good with
- Python: (Numpy, Scipy, OpenCV, Pytorch, Keras, Flask)
- Known programming languages:
- C# •C++ •javascript •SQL • PL / SQL • JAVA • CSS • HTML5
- Basics of:
- 3D modelling: •Blender •3DS max
- Experienced with:
- Panda 7dof arm •holonomic,nonholonomic car • Linux • Git
- Lidar • IMU • SLAM • ROS • nvidia jetson nano •raspberry pi •Arduino
- Good with
- COMPETENCES
- English language C1
- Russian language C2
- Romanian - native
- Driving license AM, B1, B
Robotics & AI projects

3D LiDAR pose-graph slam
Source Code
- Estimate the transformation between laser scans with ICP;
- Construct the keyframe pose-graph in the FrontEnd;
- Optimize the graph in the BackEnd;

Python implementation of the RGB-D SLAM
Source Code
- Detect and Track image feature;
- Perform Ransac PnP algorithm to estimate the transformation between the 3D point cloud on the current frame and their 2D pixel correspondences in the previous frame.;
- Build a pose graph based on estimated transformations;
- Update the current pose based on estimated R and t;
- Build the sparse map;
- add Bundle adjustment;
- add loop closure detection with V-BoW;

Python implementation of the Stereo PTAM (parallel tracking and mapping)
Video result, Source Code
- Multithreads Tracking, Mapping, and Loop Closing;
- Covisibility Graph (representing the relation between keyframes, mappoints and measurements);
- Local Bundle Adjustment and Pose Graph Optimization;
- Motion Model (used for pose prediction, then for reliable feature matching);
- Point Clouds and Graph visualization;

Master thesis: Stereo-Camera–LiDAR Calibration for Autonomous Driving
Video result, Source Code
Master thesis project realized in partnership with Aalto University and Finnish Geospatial Research Institute of Finland.
- Mono and stereo camera calibration
- Camera-LIDAR synchronisation
- Stereo-camera based and LiDAR based 3D reconstruction
- Camera-LiDAR extrinsic calibration
- Sensor occlusion handling

Pong game by Panda robots, Robotics AI Lab
Video result, Source Code
- 2D to 3D projection
- Forward/Inverse Kinematics with multiple objectives by stacking Jacobians
- Path planning with Potential Fields
- Image-based and Position-based visual servoing

Autonomous driving by behavioural cloning
Video result, Source Code
- Run Udacity car simulation & record some data (images, steering angle, velocity, etc)
- Data preparation - balance the dataset, augment data
- Train a regression model (input images, output steering angle)
- Use the model to control the car steering angle

Small scale autonomous driving car
Simulation,
Real car,
Source Code
- Traffic line detection (OpenCV hough transform) & following
- Traffic sign detection & recognition (CNN model)
- SLAM with EKF for mapping & localization
- 2D spline function for path planning
- Stanley steering and PID controller for path tracking

Reinforcement learning, Agent learns to play pong game, from pixels
Source Code
- The agent is the green paddle
- Pass the image through CNN to extract the features
- Used fully-connected layers to learn the behaviour
- Used softmax to output 3 actions probabilities (up, down, stay)
- Policy gradient optimize the weights to maximize the reward

- Detect feature descriptors (harris corner, SIFT)
- Track the features in 2 consecutive images, using optical flow.
- Estimate Essential matrix (E) between points correspondences, using RANSAC model fitting.
- Estimate camera rotation (R) and translation (t) based on essential matrix.
- Update the current pose based on R and t.
Other projects

Stock-Price-Prediction
Source Code
- Time series prediction
- The application allows you to upload the current stock prices, and it predicts the future prices
- It allows the user to train a new model on its own data
- The app can be automated completely, the agent will buy and sell stocks independently.
- There are LSTM models implemented in Keras, moving average, RL agents.

Self Hotel - Web app
Source Code
This website provides room registration, self check-In and check-Out services.
- The app is based on:
- client-side: - Javascript(Bootstrap & jQuery), HTML5,CSS.
- server-side: - C#, .NET, sql Server.
- SQL:
- Data description language (DDL).
- CRUD Operations (create, select, update, delete)