projects
A curated collection of my work—from Research initiatives to course-based assignments. Click on any card for full details.
Research Projects
Flexure-Hinge Torque-Sensor Study — Increasing Moment Capacity of 6-DoF FT Sensors
Lightweight flexure stage increases 6-DoF sensor moment capacity to ≥280 Nm while preserving 1 N force resolution. Analytical, numerical and experimental investigation of crossed-strip flexure pivots to decouple force and moment limits in 6-DoF force-torque sensors. Conducted at TUM-MIRMI under supervision of Mehmet Can Yıldırım and Prof. Sami Haddadin.
mechanics robotics sensing flexure
Passive Walker RL — Curriculum-Driven Biped Locomotion in JAX & Brax
A three-stage pipeline (≤ 300 LoC per stage) that bootstraps a passive-dynamic biped from a finite-state expert to a GPU-scale PPO policy in minutes. MuJoCo supplies fidelity; Brax pushes > 1 M env-steps s⁻¹ for massive sweeps, yielding smooth, sample-efficient walking with a single 1 M-param network.
reinforcement-learning robotics jax brax
Panda-Base Digital Twin — Modular Simulation of a High-Moment Manipulation Setup
Modular Gazebo-based digital twin of a Franka Emika Panda arm mounted on a custom aluminium-profile base. Developed during a TUM-MIRMI research internship, it enables controller tuning and moment-load evaluation entirely in simulation.
robotics ROS Gazebo simulation
DiffSwing: Neural Cart-Pole Control with JAX & MuJoCo
A hybrid control system that combines neural network energy shaping with classical LQR stabilization for cart-pole swing-up. Trained end-to-end in differentiable JAX simulation with real-time MuJoCo deployment achieving 98% success rate and 1.9s swing-up time.
control reinforcement-learning robotics neural-networks jax
How can a robot place an item on a cluttered desk?
PAGE WİLL BE UPDATED
Multiexplorer — Multi-Robot Frontier Exploration with TurtleBot3
My first ROS project: a Gazebo-based study comparing single-, double-, and triple-robot frontier exploration, SLAM, and map-merging with TurtleBot3.
robotics ROS Gazebo SLAM multi-robotCourse Projects
Hand-Steer Sim — Real-Time Gesture Teleoperation for Mobile Robots
Hand-Steer Sim is a webcam-driven teleoperation stack that translates hand gestures into ROS /cmd_vel commands in real time. MediaPipe landmarks feed a 1 k-param MLP for four static actions and a 6 k-param LSTM for steering trajectories, achieving 99 % accuracy with 13 ms end-to-end latency. One launch or Docker run controls a Gazebo (or real) differential-drive robot—no joystick required....
robotics computer-vision HRI
3RRP Mechanism — Kinematic & Dynamic Analysis
A full symbolic + numerical study of a 3-RRP planar parallel manipulator: closed-form kinematics, workspace / isotropy metrics, and Kane- vs-Lagrange dynamics validated in Simulink.
robotics kinematics dynamics simulation
2-DOF Manipulator — Model-Based Real-Time Control
A MATLAB/Simulink + TI C2000 control stack that combines computed-torque, PD + gravity compensation, and joint/task-space transformations to track millimetre-level trajectories on a 5-link planar robot in real time.
robotics control real-time
Planar Elbow Manipulator — Design & Control
A 2-DoF planar elbow manipulator that uniquely met every mechanical requirement in ME408, integrating robust CAD, power-electronics, and dual-loop control for precise trajectory tracking.
mechanical-design power-electronics control-systems robotics
FPGA-AirHockey — Digital Showdown on Nexys A7
Two-player air-hockey game on a Nexys A7 FPGA; a Verilog state-machine drives a 5×5 LED rink and real-time score displays.
fpga verilog embedded-systems game