Stepan Feduniak

I work on reinforcement learning for Vision-Language-Action (VLA) models at the Learning Systems and Robotics (LSY) lab at the Technical University of Munich (TUM).

Alongside this work, I am a first-year Bachelor student in Mathematics at TUM.

I have a strong interest in the intersection of mathematics and robotics, particularly in enabling the robot learning from real-world interaction.

Email      GitHub      LinkedIn      X (twitter)      CV

Researcher, Learning Systems and Robotics (LSY) lab, TUM

Nov 2025 - Present Technical University of Munich

Real World Reinforcement Learning for Vision-Language-Action (VLA) models.

B.Sc. Mathematics, Technical University of Munich

Oct 2025 - Present Technical University of Munich

First-year bachelor student in Mathematics.

RECAP: Pi0.6 Implementation for LeRobot

Feb 2026 GitHub Repository / Open Source

Implementation of the RECAP method from the Pi0.6 paper into the LeRobot.

Humanoid Manipulation Hackathon (Winner)

Sept 2025 Siemens Challenge Winner / LeRobot

Won the Siemens challenge by implementing BESO and explainability for VLA for robust fuse mounting. [Read more on LinkedIn]

Research Assistant, Intuitive Robots Lab (IRL), KIT

May 2025 - Oct 2025 Karlsruhe Institute of Technology (KIT)

Worked on integrating tactile sensing into flow-matching and Vision-Language-Action (VLA) policies with hierarchical models.

  • Fine-tuned and deployed VLA models (GrootN1, SmolVLA, FlowerVLA, Pi0) on a Franka Emika Panda robot.
  • Implemented multi-modal sensor fusion approaches.
  • Analyzed tactile embeddings from different models.

Olympiad Mathematics Blog & Tutoring

Apr 2024 - Sep 2025 Lviv, Ukraine

Founded a math club and led a volunteer math Olympiad preparation tutoring team.

Self-Learning Resources

Ongoing

A list of resources I've found most useful for learning Deep Learning, Reinforcement Learning, and Robotics.