Three models
Coordinated control of multiple robots; intelligence transferable between machines
Google's Gemini Robotics 2: Multi-Robot Coordination and Rapid Adaptation to New Hardware
Google DeepMind announced Gemini Robotics 2, an AI suite for robot control, demonstrating fully autonomous whole-body movement on Apptronik's Apollo 2 humanoid. Two features are central. First, the reasoning model ER 2 assigns work across multiple robots and coordinates their operation. Second, the system adapts to a different robot with fewer than 200 training examples, collected in a few hours.
Deep Dive — Key Points
- Three models. A vision-language-action model for movement, ER 2 for planning and coordination, and On-Device 2 for local execution on the robot. ER 2 is available through Google AI Studio.
- Natural-language operation. From an instruction such as "put the watering can into the green bin on the bottom shelf," the robot executes walking, grasping and placement autonomously.
- Multi-robot coordination. ER 2 distributes work across different robot types and manages minutes-long tasks, including progress monitoring and re-execution after failures.
- Safety functions. A benchmark (ASIMOV-Agentic) evaluating refusal of unsafe instructions and requests for human intervention was released alongside. Detection of nearby people, with safe stopping, was also strengthened.
This AI covers coordination between robots. Allocating work across an entire facility — orders, inventory, personnel — is handled by the software layer above it (the logistics OS / WES), whose importance increases as the intelligence layer becomes hardware-agnostic.
Source: Robotics & Automation News Jul 31, 2026
