Imagine asking a household robot to tidy a messy living room, then watching it decide how to move, bend, grab objects, and recover when something slips. Google is working toward that kind of smart home.
Google DeepMind announced Gemini Robotics 2 and Gemini Robotics ER 2 on July 30, 2026. The technology is not a retail robot, but a foundation model stack designed to help manufacturers build more capable machines.
For homeowners interested in what comes after voice assistants, the shift is significant. Instead of smart devices simply responding to commands, future robots could potentially see a room, understand an instruction, and perform several physical actions.
Google is giving robots a body
Gemini Robotics 2 is not a robot that consumers can order. It is part of a family of AI models designed to work with different robot embodiments, including full humanoids and bi-arm platforms.
The Gemini Robotics 2 vision-language-action model converts visual and language inputs into motor control signals and can direct whole-body humanoid movements from the feet to the fingertips.
Google pairs the action model with Gemini Robotics ER 2 for higher-level reasoning, while real robot deployments still rely on hardware-specific control and safety systems for functions such as collision-free motion, balance, and force control.
How could a robot handle chores?
The process resembles human planning, although the underlying technology is considerably more complex. Cameras mounted around a robot provide visual information, while the AI interprets instructions and decides what to do next.
Gemini Robotics ER 2 acts as an embodied reasoning system that can break a task into steps, monitor progress through continuous video, and communicate when assistance is needed. It can also support coordination between multiple robots.
A watering can shows the concept
In one of Google DeepMind’s Gemini Robotics 2 demonstrations, Apptronik’s Apollo 2 humanoid is instructed to put a watering can into a green bin on the bottom shelf. The robot walks to the table, picks up the watering can, moves to the shelving unit, and places it in the requested destination.
That matters for household-style tasks because putting something away can require walking, bending, reaching, changing posture, and carefully manipulating an object across different locations.
What makes the system different?
Google highlights whole-body control as a major milestone. Gemini Robotics 2 can operate a full humanoid body through one policy, combining legs, torso, arms, and multi-finger hands for physical tasks.
The system can work with dexterous hands containing 22 degrees of freedom, as well as simpler parallel-jaw grippers. That flexibility could support actions ranging from tying knots to packing objects tightly.
Gemini Robotics ER 2 is also designed for longer sequences. Google says it can orchestrate tasks lasting several minutes and involving hundreds of decisions, while checking progress and handling handoffs between steps.
Why this could change smart homes
Most smart home products are still stationary. Cameras watch, sensors detect movement, speakers answer questions, and robotic vacuums cover floors, but these devices generally cannot manipulate the wider environment.
A mobile robot with dexterous hands could potentially tidy shelves, organize bins, move laundry or retrieve objects from awkward places. Those tasks require perception and physical coordination that conventional home automation does not provide.
Robots could work as a team
Google is also emphasizing multi-robot collaboration. Gemini Robotics ER 2 can give different robots a shared understanding of a task, allowing machines with different physical abilities to coordinate their actions.
That could eventually connect a humanoid with a wheeled robot or other specialized machines. One device might carry an item while another handles a task requiring a different shape, reach, or movement capability.
Local AI could make robots faster
Gemini Robotics On-Device 2 is designed to run directly on a robot rather than relying on a cloud round-trip for every action. That could matter when a machine needs to respond quickly around people.
Google says the model can adapt to a new bi-arm robot in a few hours using fewer than 200 demonstrations. Faster adaptation could help manufacturers customize robotic intelligence for different hardware and household environments.
Little-known fact: Google’s original on-device robotics model (a predecessor to On-Device 2) was unveiled back on June 24, 2025, and could already fold clothes and unzip bags with no internet connection.
Safety remains a major hurdle
A household robot would operate much closer to people than many industrial machines, making safe behavior essential. Google says its robotics work includes improved human proximity detection and safety constraint following.
The system can stop when a person moves too close and resume when the area becomes clear. Google has also introduced the ASIMOV-Agentic benchmark to evaluate uncertainty handling, safe tool orchestration, and responses to nearby people.
These measures are important, but they do not make household robotics risk-free. Real homes contain children, pets, clutter, fragile objects, and constantly changing conditions that can challenge both hardware and software.
This is still far from a product
The biggest caveat for homeowners is availability. Gemini Robotics 2 is not a consumer robot available at retail.
Gemini Robotics ER 2 is publicly available to developers through the Gemini API and Google AI Studio, while the Gemini Robotics 2 vision-language-action model and Gemini Robotics On-Device 2 remain limited to early-access or trusted-tester programs.
Physical performance also depends on the robot’s embodiment. Google reported medium-to-high success on whole-body and gripper-based dexterous tasks while describing multi-finger dexterity as still challenging.
Little-known fact: In Google’s own tests, the system actually performed more reliably with simple two-finger grippers than with complex multi-fingered hands.
How Google differs from earlier efforts
Google’s earlier RT-2 work showed how vision-language models could be transformed into vision-language-action models for robot control, with demonstrations centered largely on manipulation using robot arms and mobile manipulators.
Gemini Robotics 2 extends that approach to learned whole-body humanoid control, allowing a robot to coordinate actions such as walking, crouching, reaching, and manipulating objects. Hardware-specific low-level controllers remain important for functions such as collision-free motion, balance, and force control.
That distinction could matter in homes, where useful tasks may involve objects spread across different rooms and heights. Google’s broader robotics strategy is also designed to transfer learned capabilities across multiple robot embodiments rather than remain tied to one hardware form.
What homeowners should watch next?
The most interesting development may be the transition from smart devices that understand commands to machines that understand tasks. That is a much harder problem because the robot must interpret its surroundings and make physical decisions.
Speed remains a limitation, with current robots moving cautiously rather than at a human pace. Hardware quality, battery life, reliability, and safe interaction will also determine whether these systems become genuinely useful in homes.
Still, Gemini Robotics 2 points toward a different definition of smart home automation. The next major upgrade may not be another sensor or voice feature, but a mobile agent capable of seeing, planning, and physically helping.
The smart home may become active
Google’s robotics work suggests household automation could eventually extend beyond controlling lights, thermostats, and appliances. A capable robot could connect those digital systems with physical actions inside the home.
If those capabilities mature, household robots could become a new layer of smart home infrastructure. Instead of asking whether a home can respond to a command, homeowners may eventually ask whether it can understand a goal and accomplish it safely.
TL;DR
- Gemini Robotics 2 gives compatible robots learned whole-body control, allowing humanoids to combine locomotion, reaching, and dexterous manipulation while working with hardware-specific control and safety systems.
- Gemini Robotics ER 2 can plan multi-step tasks, monitor progress, and coordinate different robots, pointing toward more capable physical automation.
- Google’s technology is not a consumer household robot. Gemini Robotics ER 2 is available to developers, while the vision-language-action and on-device models remain in early access or trusted tester programs.
- Gemini Robotics On-Device 2 can run locally and adapt to new bi-arm robot embodiments with fewer than 200 examples, although sensors, actuators, controllers, and mechanical design remain important constraints.
- Future smart homes could eventually shift from primarily stationary connected devices toward mobile robots that understand goals and perform physical tasks.
This article was made with AI assistance and human editing.
If you liked this, you might also like:
