Google DeepMind's AI Model Now Takes Control of Entire Humanoid Robots

Google DeepMind has upgraded its Gemini Robotics AI model to control entire humanoid robots. The new version focuses on whole-body motions from feet to fingertips, whereas the previous one only handled upper body movements.

With this update, robots can perform a wider range of actions, including walking, crouching, stretching, and manipulating objects. Videos show Apptronik’s Apollo 2 robot picking up a watering can or finding items on a shelf. The demonstrations are impressive and indicate that these AI tools for businesses could be useful in various industries.

Google DeepMind says the new model is a crucial step towards enabling robots to complete complex real-world tasks, which require whole-body coordination. While movement speed still needs improvement, this update marks a significant advancement in their capabilities.

The Gemini Robotics 2 model also boasts improved dexterity, allowing robots to control five-fingered hands and perform precise actions like sealing a Ziploc or unscrewing a lightbulb. This level of precision is exactly what businesses need for complex tasks that demand human-like manipulation.

Google DeepMind has updated its Gemini Robotics ER (embodied reasoning) model as well, which helps robots analyze their surroundings, process instructions, and perform multi-step tasks. The new version can complete tasks over extended periods and understand when each task begins and ends.

One notable feature of the ER 2 update is its ability to detect humans nearby and trigger safety protocols if someone approaches too closely. This makes it the safest robotics model currently available, according to Google DeepMind. The company emphasizes that robot safety is a top priority.

Improvements have also been made to the Gemini Robotics On-Device Model, which can run on a robot without an internet connection. This updated version adapts faster to new embodiments, even those with drastically different shapes and sensors.

The ability of multiple robots to work together and complete tasks is supported by the updated models. A video shows Apptronik’s Apollo 2 instructing Google’s dual-arm robot to put tools inside a bin while cleaning the garage. This kind of collaboration could be particularly useful in industries like manufacturing or logistics.