欧博allbetApollo humanoid robot tackles unknown obje

AI and Robotics

Watch Apollo humanoid robot tackle never-before-seen objects with Google DeepMind

Gemini Robotics enables multi-embodiment control, letting robots like Apollo learn from demonstrations, follow instructions, plan tasks, and handle unfamiliar objects.

ByJijo Malayil

AI and RoboticsDec 11, 2025 08:29 AM EST

Copy or Share

Google News Preferred Source

ByJijo Malayil

Dec 11, 2025 08:29 AM EST

Copy or Share

Google News Preferred Source

Apollo humanoid robot carrying out tasks in Google lab.

Apollo humanoid robot carrying out tasks in Google lab.Google DeepMind/YouTube

Google DeepMind recently demonstrated that Apptronik’s Apollo robot can perform complex tasks using objects it had never encountered before.

In the showcase, the robot is seen handling oddly shaped or soft items, responding to verbal instructions such as “pick up the green block,” and adjusting in real time as containers or objects are moved.

The demonstration also showed Aloha robots performing daily activities such as packing lunches, sorting items, and listening to and responding to verbal instructions.

Last week, reports said DeepMind is developing an AI system with an “inner monologue,” allowing robots to observe tasks, describe them in language, and learn more efficiently.

AI powers humanoid robots

Google DeepMind’s latest robotics push signals a major step toward general-purpose humanoid machines. In partnership with Apptronik, the company is pairing its advanced Gemini 3 and Gemini Robotics AI models with the Apollo humanoid robot, aiming to create a single system capable of performing a wide range of real-world tasks without retraining for each new body or environment, reports Forbes.

Gemini Robotics is built for “multi-embodiment” control, allowing it to operate everything from dual-arm industrial robots to full humanoids like Apollo. The result is a robot that can watch demonstrations, understand natural-language instructions, plan multi-step procedures, and interact with unfamiliar objects. Recent lab demos show Apollo performing everyday activities such as packing lunches, sorting laundry, placing items in containers, and navigating the kind of unpredictable settings typical of human homes.

This momentum reflects rapid advances in both AI and hardware over the past two years, driven by better models and cheaper components. DeepMind says future breakthroughs must focus on data efficiency, since training robots still requires enormous volumes of interaction and manipulation data. Safety also remains a central challenge, especially for machines meant to operate around people, pets, and fragile household items.

While today’s robots still lack the finesse needed for tasks like sealing a Ziploc bag, the fusion of Apollo’s hardware with DeepMind’s increasingly capable AI marks a meaningful shift. If the technology proves reliable outside the lab, it could push the field closer to a universal robot worker—one that understands instructions, adapts on the fly, and performs physical tasks with far fewer limitations than before, reports Forbes.

DeepMind teaches thought

DeepMind is developing a new technique that could transform how robots learn by giving them an “inner monologue.” A recent patent filing outlines a system in which AI agents watch videos or images of a person performing a task and generate natural-language descriptions of what they observe.

For example, a robot might watch someone lift a cup and internally describe the action as “the person picks up the cup.” By pairing visual input with this inner speech, the system helps robots understand actions more clearly and remember how to perform them when faced with similar situations.

The approach enables “zero-shot” learning, allowing robots to handle unfamiliar objects or tasks without prior training. DeepMind also says this method reduces the memory and processing demands typically required for robotic training, making it more efficient than existing techniques.

This work expands on DeepMind’s broader robotics advancements. Recently, the company introduced “Gemini Robotics On-Device,” a compact version of its robotics foundation model designed to run directly on a robot without relying on cloud services. According to experts, combining on-device AI with inner speech gives robots essential context, improving their ability to make decisions and adapt to complex, changing environments.

欧博ABG官网-欧博官方网址-会员登入

欧博allbetApollo humanoid robot tackles unknown obje

热点文章

最新文章

友情连接