I started in Mechatronics Engineering — building swarm robots from scratch with microcontrollers, IR/PIR sensors, and inter-robot transceivers, programming CNC machines with Arduino, and working with PLC automation systems. That foundation in hardware-software integration pushed me toward AI, where I worked on computer vision pipelines, GAN-based super-resolution, and nature-inspired optimization.

Today, I am an ERASMUS MUNDUS IFRoS scholar (UdG, Spain & ELTE, Hungary) — one of the most competitive fully-funded robotics master's programmes in Europe. My current work at EUROKNOWS, Hungary proposes and evaluates three progressively complex semantic mapping pipelines for mobile robotics — integrating object detection, segmentation, and vision–language models for context-aware scene understanding and natural-language interpretation in real-world environments.

Across this journey I have implemented systems end-to-end: from bare-metal embedded code to deep learning training pipelines, from kinematic controllers to EKF-SLAM, from path planners to speech-driven robot navigation — always on real or simulated hardware, always in ROS.

• Autonomous field robots — swarm intelligence (PSO), embedded microcontrollers, IR/PIR sensing, inter-robot transceiver communication
• CNC & automation systems — Arduino-based X-Y plotter, PLC programming with CoDeSys, industrial robot manipulation with CIROS
• Deep learning models for super-resolution and object detection
• Nature-inspired optimization frameworks for urban infrastructure planning using GIS data
• Perception & Navigation pipelines for mobile robots — RGB-D, LiDAR, ArUco, EKF-SLAM
• Kinematic controllers for mobile manipulators using task-priority redundancy resolution
• Kinodynamic planners combining RRT, spline smoothing, and Pure Pursuit control
• Vision–Language systems for semantic scene understanding and instruction grounding

A mobile robot that receives spoken commands ("go to the red chair"), detects and localizes the object in 3D using RGB-D + YOLO, and navigates safely — with active frontier-based search if the object is not visible. Built on AgileX SCOUT Mini with ROS, running end-to-end in real-time.

Full pick-and-place system on TurtleBot2 + uArm Swift Pro using hierarchical null-space control, ArUco marker detection, and dead reckoning. 11 sequentially ordered tasks executed autonomously in Stonefish simulation.

Lightweight SLAM fusing odometry, IMU, and bearing-only ArUco observations. Demonstrated robust re-localization after deliberate high-speed drift and wall collision — map orientation remained accurate throughout.

RRT + tensioned B-spline smoothing + Pure Pursuit controller for continuous forward-motion exploration of unknown maps. Achieved full coverage in 65.7% of 35 simulation runs with under 1 forced stop per run on average.

AI-driven optimization framework for urban rail network planning integrating GIS spatial data with nature-inspired algorithms (PSO, GA). Evaluated across real urban landscape datasets and published as a Taylor & Francis book chapter.

GAN-based super-resolution architecture benchmarked against existing works on aerial military imagery datasets — outperformed state-of-the-art on PSNR/SSIM metrics.

5 autonomous field robots with PSO-based swarm intelligence, IR/PIR sensing, and transceiver-based inter-robot communication. Bachelor's dissertation — state-funded research grant recipient.

• Generation of Super-Resolved Images Using Deep Neural Networks — IEEE ICCSC 2024
• Advanced AI Strategy for Urban Rail Network Design using Nature-Inspired Algorithms — Taylor & Francis 2024 · arXiv
• AI Based Navigation in Quasi Structured Environment — arXiv 2023