Gemini Robotics 2 brings whole body intelligence to robots

State of the technology

• Many commenters ask where this really stands: how often it succeeds, how much instrumentation is needed, and how well it handles doorknobs, falls, tight spaces, etc.
• Reported task success is ~60% overall, ~80% accuracy on some benchmarks; some tasks (e.g., screwing in a light bulb) are notably weak.
• Practitioners say it’s “not there yet” for unsupervised practical use, especially in homes; humanoids are mechanically complex and maintenance‑heavy.
• Dexterity remains a bottleneck: current grippers lack human‑like sensing and control, and fine manipulation is still hard.

Demos, hype, and comparisons to self‑driving

• Several comments liken this to LLM progress: robotics might be at a “GPT‑1” stage with a “GPT‑2 moment” coming, though others say progress from Gemini Robotics 1→2 is small versus GPT‑2→3.
• Strong skepticism about polished demos: reports of staged tasks, narrow happy paths, and even teleoperation being passed off as autonomy.
• Self‑driving is a frequent analogy: visible progress (e.g., Waymo), but rollout is slow; some think humanoids will follow a similar long “slow burn”.

Use cases: home, industry, and care

• Home cleaning/cooking/folding laundry is the most discussed aspiration. Many would pay car‑level prices if robots could reliably handle chores.
• Others argue specialized devices (Roomba, dishwashers, “laundry folder cubes”) are more realistic than full humanoid maids.
• Industrial uses (factories, warehouses, hotels, restaurants) are seen as nearer‑term: easier ROI and more controlled environments.
• Elder care is a major hoped‑for application, though some find outsourcing intimate care to robots emotionally disturbing.

Economics and adoption

• For businesses, threshold is “cheaper than 3 shifts of humans”; for households, anything above a few thousand dollars is a hard sell except for affluent users.
• Many expect leasing/subscriptions rather than ownership; concerns about lock‑in and remote bricking.
• Some argue initial market will be wealthy households and firms, exacerbating inequality.

Humanoid vs specialized robots

• One camp: humanoids fit existing human‑designed spaces and can be dropped into current workflows.
• Other camp: humanoids are an inefficient gimmick; task‑specific robots (industrial arms, Roombas, tethered cubes) already deliver value and will dominate.

Safety, reliability, and privacy

• Big worry: a robot strong enough to be useful is strong enough to seriously injure or cause huge property damage, even without “rogue AI”.
• Current systems tend to stop when humans get too close; commenters note robots for close physical interaction will need another level of safety engineering.
• Privacy concerns: household robots will be packed with sensors, likely cloud‑connected, becoming powerful surveillance devices.
• Some insist on fully local models and hardware kill‑switches; others doubt that’s compatible with current compute needs and business models.

AI models and control architectures

• Discussion around vision‑language‑action (VLA) models vs alternative approaches (e.g., world‑model JEPA).
• LLM/VLA latency is a concern: 10 Hz control is far slower than low‑level controllers, so many argue for hybrid hierarchies (fast PID‑style control under slower high‑level planning).
• Debate over whether hardware (actuators, sensors) or software (learning and data) is the real bottleneck; some point to rapid recent actuator advances, others see little change since early humanoids.

Societal and ethical implications

• Strong concern that once inference + hardware are cheaper than labor, capital owners will no longer “need” human workers, driving unemployment and wealth concentration.
• Optimists imagine a world where people work only by choice; critics see a more likely outcome of elites owning robots and using them for profit, policing, warfare, and social control.
• Several comments stress that outcomes hinge on political and economic decisions (democratizing capital vs concentration), not just the technology.