Google DeepMind taught Apptronik’s Apollo 2 humanoid robot to play cornhole by analyzing a single video of a human throwing a bag. The Gemini 4 Argon model broke down the human motion, translated it to the robot’s own joints, then practiced in a MuJoCo simulation before executing the throw in the real world. The robot landed seven consecutive shots. This Video-to-Sim-to-Real pipeline cuts modeling time from weeks to hours and scored 87 % on ERQA, an embodied reasoning benchmark. This breakthrough could benefit logistics, manufacturing, and consumer robotics sectors, though the demonstration remains a simple, controlled case.
Source: Read the original article
Disclaimer: this content is for information purposes only and is not financial advice. Cryptocurrencies are highly volatile: you may lose all of your capital. Always do your own research. Legal notice
