Talk
The talk has three blocks and a live robot demo.
Block one - why robots need Python and what ROS 2 is (~10 min). Why robotics needs middleware, what ROS 2 brings to the table, and why its architecture will feel familiar to any Python developer: nodes are microservices, topics are a message broker, services are RPC. We dig into rclpy - the first-class Python client library on top of the C core - and recent improvements in ROS 2 Kilted Kaiju (2025): extended type-checking support, the events executor with up to 10× speedup, and the ament_mypy tool.
Block two - ecosystem: what a Python-powered robot is made of (~8 min). A quick tour of the key libraries: Nav2 for navigation, MoveIt 2 with Python bindings for manipulation, URDF for robot description, Gazebo for simulation, cv_bridge for camera data. For each area - a minimal Python snippet so the audience can see what the code actually looks like.
Block three - simulation and RL: training a robot without buying one (~10 min). The simulator
landscape (Gazebo, MuJoCo, NVIDIA Isaac Lab), the Python RL stack (Gymnasium, Stable-Baselines3),
and the sim-to-real transfer pipeline. We walk the path from gym.make() to a trained manipulation
policy.
The finale is a live demo: a small robot on stage controlled by a ROS 2 node written in Python right
there in the slides. So the audience can see that the code we've been discussing for thirty minutes
actually moves real hardware.
About the Speaker
I'm a software engineer with about twenty years of experience in product development, architecture, and team building. I spent many years at Yandex, where I helped create the first versions of Yandex Disk and Yandex Music, and later led development teams across Mail, Disk, and Passport.
After that I moved into startups and venture building - projects at the intersection of software, data, and messy real-world systems. I worked on home blood analysis, legal entity scoring, and race management systems for autonomous vehicles. At Yandex Armenia I was part of the team behind a delivery route optimization product - the kind of problem where algorithms meet real roads, real drivers, and real traffic.
Today I lead the development of DeepAgent, Yandex's internal deep research AI product. My day job is LLMs, agent workflows, ML pipelines, and the practical side of making AI systems that actually work in production.
My hobby is robots. I experiment with computer vision, world models, and robot learning - and run a Telegram channel about robotics and AI agents (t.me/robots_of_dawn). I organize IT Breakfast - a developer meetup series in Yerevan - and occasionally bring a robot to a talk to prove the code on the slides does something real. Based in Yerevan, Armenia.