New DeepMind researcher says Argon lets a humanoid learn cornhole from video
A researcher announcing a move to Google DeepMind says Gemini 4 Argon used reasoning, coding and multimodal physics understanding to let a humanoid learn consistent cornhole shots by watching a human video.
What a day to announce that I've joined @GoogleDeepMind as a Research Scientist! 🚀 We are witnessing a paradigm shift in Physical AI: Agentic Robotics. Learning skills directly from watching a human video has been a holy grail of robotics. Seeing Gemini 4 Argon unlock this through sheer reasoning, coding and multimodal physics understanding was surreal. Sinking cornhole shots on a humanoid takes control over fluid whole-body dynamics and pinpoint accuracy - I can't even hit this anywhere near that consistently! You give Argon a video of a person doing it, let it cook and boom: it's draining shots like its nobody's business. I'm super thrilled to be pushing towards Physical AGI with this incredibly talented team and leadership at this unprecedented time for robotics - the pace of progress is going to be insane! And the craziest part? This was just v0. We are barely scratching the surface of what these models are capable of doing. Buckle up 👀🤖
This post quotes another post: x.com/google/status/2110041301253075358