Gemini 4 Argon learns cornhole from video and sinks seven shots on a humanoid robot
A post says Argon completed a Video-to-Sim-to-Real pipeline: watching a human play cornhole, computing whole-body joint trajectories, generating simulation code, and sinking seven straight shots on Apptronik's Apollo.
Gemini 4 Argon just did something concrete on a humanoid: it watched a human playing cornhole, mapped the environment, calculated whole-body joint trajectories, generated the simulation code, practiced in sim, then transferred to Apptronik’s Apollo and sank seven shots in a row.（Video-to-Sim-to-Real） Argon is a general multimodal model, yet its physical reasoning is already strong enough to handle multi-view video, contact mechanics, and proprioception. Together with Gemini Robotics (VLA for action) and Gemini Robotics ER (embodied reasoning for planning and orchestration), the stack is getting cleaner. General models are developing a deeper understanding of humanoid robots and the physical world.
This post quotes another post: “Gemini 4 Argon learns from video to teach Apollo robot 7 straight cornhole throws”