Mustafa Sepen Robotics Software Engineer

Robotics Software Engineer

Mustafa
Sepen

  • Autonomous Navigation
  • Localization
  • ROS 2
  • GNSS/RTK
  • Embedded Systems

Outdoor autonomy on real hardware — RTK GNSS localization, Nav2 waypoint navigation and the embedded motor interface underneath, running today on three ground vehicles that share one ROS 2 stack.

Currently
Robotics Engineer, Swegeo
Platforms
Tracked · 4×4 · Ackermann · USV
Primary stack
ROS 2 Humble · C++ · Python
Based
Istanbul, Türkiye

Selected work

Motion Detection — CUDA / YOLO — read the case study

Optical flow, GPU processing and learned detection

Classical CPU
Farnebäck flow, temporal masking, region markers
CUDA / OpenGL
GPU flow and experimental component kernels
YOLO11
Fine-tuned detector with library tracking

Three separate implementations of the same problem — classical optical flow, a CUDA/OpenGL path, and a fine-tuned YOLO11 detector — built to compare their failure modes rather than their frame rates.

Experience

  1. Sep 2025 — Present

    Robotics Engineer

    Swegeo

    Autonomous ground vehicles built from the electronic architecture up: ESP32 drive control under a Raspberry Pi 5 running ROS 2 Humble, and a custom ROS 2 driver for a dual-antenna RTK GNSS receiver with an integrated NTRIP client.

    Outdoor waypoint navigation with Nav2 over EKF sensor fusion, plus the browser-based mission console used to plan and supervise runs in the field.

  2. Feb 2025 — Jun 2025

    Simulation Systems — R&D

    Baykar

    Moving-object detection for UAV footage, from OpenCV baselines to hardware-accelerated real-time processing with CUDA.

    Integrated YOLOv11 for complex scenes and ran an R&D comparison of the approaches on performance and accuracy.

  3. Aug 2025

    Autonomous Systems — Software Team

    Team Navi · Njord Challenge, NTNU

    ROS 2 autonomy for an unmanned surface vessel: obstacle mapping, mission state machines, and dock approach and exit logic.

    4th place among 11 international teams, Trondheim.

  4. Oct 2024

    Autonomous Systems — Software

    Teknofest Unmanned Surface Vehicle Competition

    Perception with a ZED camera and YOLO, A* route planning and Pure Pursuit path tracking across the autonomous vehicle stack.

    Finalist. Most Original Software Award.

  5. 2020 — 2025

    B.Sc. Computer Engineering

    Bahçeşehir University

    Full scholarship. Honour degree, GPA 3.15 / 4.00.

Engineering focus

Autonomy

  • ROS 2 Humble
  • Nav2
  • Waypoint navigation
  • Regulated Pure Pursuit
  • Behaviour trees
  • Command arbitration

Localization

  • RTK GNSS
  • Dual-antenna heading
  • NTRIP corrections
  • robot_localization
  • Dual EKF
  • TF frame topology

Embedded

  • ESP32-S3
  • STM32H7
  • micro-ROS
  • FreeRTOS
  • Ethernet / UDP
  • Motor control and odometry

Perception

  • OpenCV
  • YOLO11
  • ZED stereo depth
  • CUDA
  • LiDAR integration
  • Optical flow

About

Istanbul, Türkiye
Available for robotics roles in Europe

I work at the boundary between robotics software and the vehicle itself — localization and navigation on one side, embedded control and physical integration on the other.

In practice that means writing the ROS 2 stack, writing the firmware it talks to, wiring the GNSS receiver and the IMU, and then taking the vehicle outside to find out what the configuration actually does. Most of what I know about EKF tuning, frame conventions and command arbitration came from watching a real platform get it wrong in a field.

The platforms so far are ground and marine: a tracked skid-steer UGV, a 4×4 differential rover and an Ackermann RC crawler running one shared autonomy architecture, and an unmanned surface vessel built with a team for the Njord Challenge at NTNU. I write C++ and Python, and I am comfortable being the person who owns the part where the software meets the hardware.