If a humanoid robot succeeds 32% of the time at sweeping, is that a failure — or the Android moment?
Google DeepMind’s new robot brain runs a whole humanoid from one model but publishes success rates as low as 32% on household tasks like using a dustpan. Is a robot that drops a third of what it picks up a sign the humanoid hype is oversold, or exactly what an early general-purpose platform is supposed to look like before it compounds?
Commentaires (1)
In the July 31, 2026 episode of Minds, Bodies, and Terawatts, we dug into Gemini Robotics 2 — and the most interesting thing about it wasn’t the demo video, it was the technical report where DeepMind published its own failure rates. There’s a real argument that publishing 32% is the most credible thing a robotics lab has done all year, and that the number matters less than the fact that one learned policy now drives legs and fingers together instead of two systems handing off. The other half of the story is the business model: Google isn’t building the robot, it’s shipping the brain onto Apptronik’s and Boston Dynamics’ hardware, which is the Android playbook applied to bodies. Whether that compounds fast or stalls is the question worth arguing about. Listen to the episode, then tell us where you’d put your bet.
Related reading on unscarcity.ai:
Envie d'aller plus loin ?
Obtenez le plan complet dans <em>L'ère de la post-pénurie : Repenser la société à l'ère des machines</em>
