Google has announced a new model, "Gemini Robotics-ER 1.6," aimed at enhancing the "embodied reasoning" capabilities of robots.
Model ReleasesGoogleGemini Robotics-ER 1.6
Google Launches Gemini Robotics-ER 1.6 Inference Model for Robots via API
This article is a translation. Read the Japanese original
The company stated that for robots to become truly useful in daily and industrial settings, they must be able to reason about the physical world rather than simply following instructions.
The new model specifically strengthens capabilities such as visual and spatial understanding, task planning, and success determination.
It is reported that accuracy in pointing, counting, and success determination has improved significantly compared to the previous 1.5 and Gemini 3.0 Flash models.
Additionally, through joint development with partner Boston Dynamics, a new feature for reading complex gauges and scales (instrument reading) has been added.
This model functions as a high-level reasoning engine for robots and can natively call tools such as Google Search and VLA (Vision-Language-Action) models.
For developers, the model has become available through the Gemini API and Google AI Studio. Colab notebooks containing configuration and prompt examples have also been released.
Source: Gemini Robotics-ER 1.6 (HN 219pt, 84 comments) (HN Search (backfill), 2026-04-15)