Upgrade to SI Premium - Free Trial

Google launches Gemini Robotics ER 2 model for robot task planning

July 30, 2026 11:02 AM

Google has released Gemini Robotics ER 2, an updated embodied reasoning model designed to serve as a high-level planning system for robots, according to a company statement dated July 30, 2026.

The model is available to developers through the Gemini API and Google AI Studio, and is in private preview on the Gemini Enterprise Agent Platform. It integrates with the Gemini Live API using a bidirectional streaming endpoint intended to reduce latency in robot control tasks.

Gemini Robotics ER 2 is designed to handle conversation, physical world understanding, and multi-step task planning, delegating motor execution to lower-level vision-language-action (VLA) models. It can call external tools such as Google Search or user-defined functions while simultaneously executing actions.

Google states the model achieves 57.4% accuracy on progress classification tasks and 91.3% accuracy on moment-finding tasks, with a mean absolute distance of 0.96 seconds. The company says it outperforms its predecessor, Gemini Robotics ER 1.6, across tool orchestration evaluations using real VLA, simulation VLA, and human teleoperation control modes.

The update introduces multi-robot collaboration, allowing different robot types to coordinate on shared tasks. Google demonstrated the feature using Apptronik's Apollo 2 humanoid and a Franka F3 Duo robotic arm. A separate demonstration using Boston Dynamics' Spot robot showed the model orchestrating navigation and manipulation commands via natural language.

Spatial reasoning capabilities were expanded to include success and failure detection from continuous video feeds, instrument reading across ten instrument types, and improved visual question answering.

On safety, Google says the model outperforms ER 1.6 on Safety Instruction Following and Human Proximity benchmarks. The company reports the model can halt a humanoid robot when a person is detected nearby and resume autonomously once the area is clear. A new benchmark for evaluating safety constraint enforcement in VLA orchestration has also been introduced, with details in a separate safety technical report.

Categories

Corporate News