Gabriel Kohanim - When LLMs Learn to Move: AI In Robotics

This project explores Vision-Language-Action (VLA) models , a class of AI systems that connect large language models directly to physical robot control. The central case study is RT-2 (Robotics Transformer 2), published by Google DeepMind in 2023, which was the first model to prove that internet-scale language pretraining could transfer to real-world robotic tasks.
The presentation also covers the current state of humanoid robotics in 2026, including the Beijing E-Town Half-Marathon (April 2026), where a humanoid robot completed a 21km course faster than any human world record autonomously.

Sources:
RT-2 Paper — arXiv 2307.15818
Google DeepMind RT-2 Blog
Google DeepMind — Shaping the Future of Advanced Robotics
Beijing Half-Marathon Robot Breaks World Record — Al Jazeera
China Leads the Humanoid Robot Race — Rest of World
China vs Tesla Robot Race — Rest of World
RT-2 Deep Dive — Toward Humanoids, Medium

Видео Gabriel Kohanim - When LLMs Learn to Move: AI In Robotics канала GabKo Programs