Google's Gemini 4 Argon trains Apptronik's Apollo to play cornhole from video

Google says its Gemini 4 Argon robotics model watched a video of a human playing cornhole, built a MuJoCo simulation, and calculated the whole-body control needed for Apptronik's Apollo 2 humanoid.

Google DeepMind says the trained skill transferred to the real robot, which sank seven consecutive shots in a real-world trial.

Google DeepMind says this Video-to-Sim-to-Real pipeline compresses weeks of manual modeling into a few hours. The claim comes from Google's own posts, and no independent evaluation of the trial is included.