AK

ROBOTICS / SLAM / COMPUTER VISION

Alican Karaomer

I build computer vision systems that turn raw sensor data into robust 3D scene understanding: visual-LiDAR-GNSS fusion, real-time SLAM/localization, and deployment-grade depth estimation.

Explore Projects Publication: arXiv:2510.06216
  • Munich, Germany
  • M.Sc. Informatics (TUM)
  • Thesis: Video World Models + 3DGS

Background Point Cloud

Shape: Knot Points: 2800 Drag anywhere to orbit · Alt + Scroll to zoom

FLIGHT LOG

Experience & Education

WORK EXPERIENCE

  1. OCT 2024 - PRESENT

    Siemens - Working Student, Localization R&D

    Driverless Railway · Munich, Germany

    Built LiDAR/GNSS/camera fusion pipelines, registration + optimization workflows, and automation tooling for large-run reproducibility.

  2. JUL 2022 - AUG 2024

    Luxoft - Software Developer (BMW AD Platform)

    Full-time & Working Student · Munich, Germany

    Refactored modules, built JSON-to-C++ code generation, and shipped dashboard + CI improvements for defect analytics.

  3. FEB 2021 - JUN 2022

    Polonom Robotics - Robotics Software Engineer

    Istanbul, Turkey

    Contributed to robotics software workflows with focus on mapping, perception engineering, and multi-UGV fleet exploration.

EDUCATION

  1. APR 2023 - MAR 2026

    Technische Universitat Munchen (TUM)

    M.Sc. Informatics

    Thesis: Video World Models + feed-forward 3DGS. Courses: Deep Learning, 3D Geometry, Autonomous Driving, Computer Vision.

  2. SEP 2017 - JUN 2022

    Yildiz Technical University

    B.Sc. Computer Engineering (GPA: 3.46/4.0)

    Istanbul, Turkey. Thesis: 2.5D belt-based exploration for mobile robots.

  3. PUBLICATION

    DropD-SLAM

    Dynamic-scene dense visual SLAM

    arXiv:2510.06216

SYSTEM CAPABILITIES

Core Skills

SLAM & MAPPING

Building real-time localization and mapping systems from LiDAR, cameras, and IMU data.

  • GTSAM
  • Factor Graphs
  • Registration/ICP
  • Sensor Fusion
  • ORB-SLAM3
  • MASt3R-SLAM

ROBOTICS

End-to-end robotics software from perception to planning and control.

  • ROS/ROS2
  • Perception
  • Path Planning (A*)
  • Frontier Exploration
  • Multi-Robot Systems
  • Gazebo

3D CV / ML

Learned 3D representations, neural rendering, and GPU-accelerated inference.

  • 3D Gaussian Splatting
  • Differentiable Rendering
  • Depth Pipelines
  • PyTorch
  • TensorRT
  • CUDA

LANGUAGES & TOOLS

Production-grade systems programming with modern build and deployment tooling.

  • C/C++
  • Python
  • Rust
  • Docker
  • CMake
  • Bazel
  • OpenCV
  • PCL
  • Eigen
  • Ceres

MISSION ARCHIVE

Projects & Publications

Selected engineering and research milestones from CV + GitHub work.

6 interactive cards

PUBLICATION

DropD-SLAM

Real-time monocular SLAM approaching RGB-D accuracy using pretrained depth and learned keypoint descriptors.

arXiv:2510.06216

Highlights: dynamic-scene handling, learned depth priors, and robust real-time monocular tracking.

SIEMENS

Railway Corridor Mapping Pipeline

Dockerized km-scale mapping fusing LiDAR + GNSS + cameras for driverless train localization with factor-graph optimization.

Highlights: calibration-aware ingestion, long-run automation, and robust optimization constraints.

AWARD

TEKNOFEST 2021 - 1st Place

Autonomous-parking motion planning using Bezier curves; SLAM with IMU and VLP-16 LiDAR in the Robotaxi competition.

Highlights: integrated planning + SLAM under competitive field constraints.

ROBOTICS

Multi-UGV Frontier Exploration

2D/2.5D frontier exploration in C++/ROS1 enabling autonomous indoor mapping with multi-robot fleets.

Highlights: distributed frontier ranking and practical multi-robot map growth.

ROBOTICS

UAV 3D Path Planning

3D A* path-planning module with volumetric cost maps for UAV tunnel/cave inspection.

Highlights: volumetric safety costs and real-flight validation workflows.

THESIS

Video World Models + 3DGS

Ongoing M.Sc. thesis on feed-forward 3D Gaussian Splatting for fast scene representation and novel-view synthesis.

Highlights: fast scene synthesis using feed-forward representations with strong view consistency targets.

kinect_fusion Unidepth-TensorRT DepthAnythingV2-TensorRT All Repositories

ESTABLISH LINK

Contact

Open for research collaborations, robotics engineering roles, and perception systems projects. Currently based in Munich, Germany.

Emailalicaank14@gmail.com LinkedInalican-karaomer GitHubalicaank arXiv2510.06216