Google DeepMind Unveils Gemini Robotics 2, Enabling Full-Body Control and Advanced Dexterity in Robots

Google DeepMind has announced the latest version of its vision-language-action (VLA) model, Gemini Robotics 2. This updated model builds upon the company’s previous work with Gemini, which demonstrated how multimodal understanding could drive real-world action. With Gemini Robotics 2, Google DeepMind is taking a significant step forward in enabling robots to reason through every movement and unlock a broad range of tasks.

The new version includes intelligent whole-body control, advanced dexterity, and multi-robot collaboration capabilities. This means that robots can now perform complex actions like walking, crouching, stretching, and manipulating objects with precision. For instance, a humanoid robot could be instructed to walk over to a cluttered room, pick up an object, and place it in a designated bin.

One of the key features of Gemini Robotics 2 is its ability to control entire robotic bodies, not just individual limbs or appendages. This is achieved through advanced algorithms that translate user intent into intelligent whole-body control. For example, when controlling Apptronik’s Apollo 2 humanoid robot, users can ask it to perform tasks like putting a watering can in the green bin on the bottom shelf.

The model has been optimized for use with various robotic devices and can adapt to new embodiments within just a few hours of data collection. This is particularly useful for applications where robots need to operate without network latency or internet connectivity, such as in industrial settings or areas with limited infrastructure.

Google DeepMind’s previous models were able to control the upper body of humanoid robots to achieve tabletop tasks. However, Gemini Robotics 2 marks a significant advancement in physical AI by expanding its capabilities to whole-body motions. This is an important step towards enabling robots to complete more complex real-world tasks that require coordination and dexterity.

The company has also introduced Gemini Robotics ER 2, which serves as the robot’s high-level brain, processing user instructions and communicating with humans. It observes the room, reasons about the steps needed to complete a task, coordinates with the VLA to carry out actions, and tracks progress until the task is done. This setup allows robots to execute complex multi-step tasks, self-correct if a step fails, and generalize to novel situations and goals.

Another key feature of Gemini Robotics ER 2 is its ability to understand when tasks begin and end, as well as pinpointing key events that occur during a task. This marks a significant improvement in progress understanding and enables robots to more reliably execute longer task sequences lasting several minutes and involving hundreds of decisions.

In addition to these advancements, Google DeepMind has introduced multi-robot collaboration capabilities. This allows different types of robots to communicate and work together to solve complex workflows that a single robot could not accomplish alone.

Google DeepMind’s commitment to safety is also evident in Gemini Robotics 2. The company has reaffirmed its dedication to ensuring the safe navigation of robots in real-world environments, particularly when collaborating with humans. To this end, it has introduced ASIMOV-Agentic, a new benchmark for agentic safety orchestration and uncertainty resolution.

ASIMOV-Agentic measures an embodied reasoning agent’s ability to refuse unsafe tool calls from a VLA and predict whether a task is possible. It also enables the agent to proactively request human intervention when uncertain. Additionally, Gemini Robotics ER 2 has been shown to be the safest robotics model to date in safety constraint following and human proximity benchmarks.

Google DeepMind’s advancements in data analysis tools for businesses are evident in its work on Gemini Robotics 2. The company is providing a comprehensive platform that enables robots to reason through every movement and unlock a broad range of tasks, making it an essential tool for industries looking to leverage AI capabilities.