01 / About

I build robots that learn to touch the world.

Computer Science researcher and practitioner, trying to close the gap between robotics theory and what actually works in the industry. My focus is reinforcement learning and control systems.

I have a long track of open-source contributions and production level robotics software at Franka Robotics, from sim-to-real reinforcement learning up to the control frameworks that ship inside the real machines.

200+GitHub stars
FrankaRobotics, Munich
RL · ControlFocus areas
02 / Selected Work

Open-source robotics, reinforcement learning and the systems that connect simulation to real hardware.

Contact-rich cable insertion demo ► Video
Featured Intrinsic AI Challenge 2026

Contact-Rich Cable Insertion (RL)

Trained a physics-aware PPO teacher in NVIDIA Isaac Sim for SFP/SC connector insertion, then distilled it into a vision based student that reaches around 80% success from RGB alone.

~99%privileged
~80%RGB only
Isaac Sim Physics-Aware PPO DINOv2 Distillation
Franka simulation server

Franka Simulation Server

58★

A high-fidelity Genesis simulation server that speaks the Franka network protocol, so you can switch between sim and hardware at 1 kHz.

GenesisNetwork ProtocolPython
Franka ROS 2 framework

Franka ROS 2 Framework

200+/day

Led the development of an industry-standard robotics control framework that is now widely adopted across research and industry.

ROS 2C++CI/CD
Deep RL grasping

Deep RL Grasping

200★ · 99%

A general deep-RL grasping model that reaches 99% accuracy on object picking, open-sourced and used by a good amount of the community.

TensorFlowPyBulletVision
NASA Space ROS competition

NASA Space ROS

5th · $1,250

Ingenuity Mars Helicopter simulation in ROS 2 Gazebo with collective and cyclic control. Finished 5th place and won a $1,250 prize.

ROS 2GazeboC++
Robotics skills platform

Robotics Skills Platform

Vision

A vision for the future of robotics programming: you demonstrate a skill to the robot, then fine-tune it and deploy.

ReactDeep Learning
4DoF 3D printed robot

4DoF 3D-Printed Robot

4 DoF

Full ROS 2 hardware interface and MoveIt packages for an open-source 3D-printed robot, to make robotics a bit more accessible.

microROSESP32MoveIt
More on GitHub
03 / Writing

Some notes on reinforcement learning, sim-to-real and the not so glamorous software that makes robots actually work. All of it lives on Medium.

Apr 2024 Controlling the UmiRobot with ROS 2, micro-ROS and MoveItA step by step guide for driving the low cost UmiRobot manipulator. There was a big interest in cheap manipulators last year, so I wrote down the whole setup. Dec 2020 Sample Efficient Robot Training in PyBullet with SACTips and tricks to get an autonomous grasping Kuka robot in PyBullet, trained sample efficiently with the SAC algorithm. My first proper deep dive into grasping.
04 / Off the Clock

What I get up to when I am away from the keyboard.

Mountain Guide Rock Climbing Basketball 3D Printing Building My Own Hardware
05 / Now
Updated June 2026 · Munich

Building production robotics software at Franka Robotics, and spending my nights on contact-rich manipulation with reinforcement learning. More and more I am convinced the real bottleneck in robotics is good software, not only bigger models, so I try to write about it more.

Currently building

A small mobile robot running on hoverboard wheels, where i drive the two motors with an ODrive controller. Right now i am getting the low level wheel control solid first, before putting any autonomy on top.

06 / Contact

Let's build something that moves.

barisyazici169@gmail.com GitHub LinkedIn
© 2026 Baris Yazici · Munich, Germany Back to top ↑