Projects

Real-Time Computer Vision Face Detection and Tracking

Developed a real-time computer vision system for face detection and tracking using the Viola–Jones framework. Implemented Haar-like feature extraction, integral images, and multi-scale sliding window search to detect faces in live video streams. Optimized detection through cascaded feature evaluation, sub-windowing, and temporal scale tracking to enable real-time performance in MATLAB.

Teaching a UR3 Robotic Arm Using Deep Q-Learning

Developed a deep reinforcement learning system to teach a UR3 robotic arm how to reach target end-effector positions without using inverse kinematics. Designed a custom ROS-based environment with collision detection and trained a Deep Q-Network (DQN) in joint space using experience replay and reward shaping. Validated the learned policy in simulation and successfully transferred it to a real UR3 robot in the lab.

Autonomous Mobile Robot Navigation and Exploration Optimization

Designed and evaluated navigation strategies for autonomous mobile robot exploration by testing goal-selection and motion-planning hypotheses. Implemented frontier-based exploration improvements, including replanning logic and pre-orientation before motion. Applied K-means clustering to frontier points for autonomous goal selection, learning and adapting this technique specifically for this project. Performance was validated through extensive simulations and real-robot experiments, achieving reduced travel distance and more efficient exploration.

MPC Control Design for Mobile Robot Trajectory Tracking

Designed and implemented multiple Model Predictive Control (MPC) strategies for trajectory tracking of a differential-drive mobile robot under constraints. Developed and compared Linear MPC, Nonlinear MPC, and Chance-Constrained MPC to handle nonlinear dynamics and external disturbances. Validated controller performance through simulations, analyzing tracking accuracy, robustness, and computational cost.

Multi-Agent Reinforcement Learning for the Onitama Board Game

Developed a multi-agent reinforcement learning system using Q-learning applied to the strategic board game Onitama. Modeled the game environment, state and action spaces, and trained competitive agents through self-play and against baseline agents. Evaluated learning performance by analyzing win-rate evolution, agent generalization, and the impact of hyperparameters on long-term strategy learning.

Haptic Rehabilitation Game for Cognitive and Motor Therapy

Designed and developed an interactive rehabilitation application for older adults with Parkinson’s and dementia using CHAI3D. The system combines maze-based spatial tasks with hand-guided interaction, providing visual and haptic feedback through a Phantom device to improve coordination, cognition, and spatial awareness. An alternative version using Leap Motion was also implemented to ensure accessibility without force-feedback hardware.

Gaze-based interface for robotic wheelchair guidance

Developed a gaze-based human–robot interface to control and assist the navigation of a robotic wheelchair for users with reduced upper-limb mobility. Integrated eye-tracking, blink detection, computer vision, and object detection to enable assisted driving, seat control, and intention interpretation. The project explores initial steps toward autonomous guidance by estimating gaze direction relative to the wheelchair and detecting user-targeted objects in the environment.

Motion Planning and Task Execution for a Mobile Manipulator (Tiago)

Developed and compared multiple motion planning algorithms (RRT, RRT*, RRT-Connect, PRM, PRM Gaussian, HF & NF1) for a mobile manipulator in a kitchen environment, benchmarking their performance to identify the most efficient planner. Integrated task and motion planning by generating XML files combining symbolic task information with geometric motion constraints.

3D-Printed Robotic Arms for Educational Programming Activities

Built and assembled three 3D-printed robotic arms for high-school students to learn robotics and programming through hands-on activities. Using open-source models as a base, I adapted, printed, and integrated the robots with servomotors so students could control their motion via joysticks and microcontrollers. This was my first experience with 3D fabrication, including assembling and calibrating an Ender 3 printer, troubleshooting print failures, and preparing the robots for classroom use.

Design of an Educational Escape Room for Environmental Noise Management

Designed, implemented, and evaluated a virtual educational Escape Room to teach Environmental Noise Management in higher technical education. The project combines gamification and problem-based learning through a realistic industrial noise case, implemented using Genially. Student feedback and survey analysis showed increased motivation, engagement, and improved understanding of complex acoustic concepts.

Data-Driven Task-Oriented Grasping for Robotic Manipulation (Research Project)

One-year PhD research project focused on data-driven, task-oriented robotic grasping under academic supervision. The work involved an extensive state-of-the-art review, problem formulation, and the design of a research framework for synthetic dataset generation, novel grasp quality metrics, and supervised learning strategies for single and dual grippers. Although not fully implemented, the project established a solid methodological foundation and a clear research roadmap for future developments in robotic manipulation.

Real-time PID Control of a Ball-in-Tube System

Implemented real-time control of a ball-in-tube platform using an Arduino. Designed and tested different control strategies, focusing on a PID controller to regulate the ball’s height using ultrasonic distance feedback and PWM fan actuation. The project provided hands-on experience in embedded control, sensor integration, and real-time execution constraints.

Page updated

Google Sites

Report abuse