Google announced today the introduction of two models, Gemini Robotics 1.5 and Gemini Robotics-ER 1.5. These models help robots actively understand their environment and generalize the completion of complex, multi-step tasks.
Model ReleasesGoogleGemini Robotics 1.5Gemini Robotics-ER 1.5
Google Announces Gemini Robotics 1.5: Realizing Physical Agents via Two Models Separating Reasoning and Action
This article is a translation. Read the Japanese original
Specifically, the architecture is designed so that Gemini Robotics-ER 1.5 acts as the "higher-level brain" to coordinate activities, while Gemini Robotics 1.5 executes the concrete actions. ER 1.5 excels in spatial understanding and logical judgment, and is capable of calling tools such as Google Search. Meanwhile, 1.5 directly executes instructed actions based on visual and language understanding.
Both models are built upon the Gemini family foundation and have been fine-tuned with different datasets according to their roles. Combining them improves versatility for longer tasks and diverse environments.
Gemini Robotics-ER 1.5 is positioned as the first reasoning model optimized for embodied reasoning. It is reported to have achieved state-of-the-art performance compared to similar models across 15 academic benchmarks, including ERQA and Point-Bench.
Regarding availability, Gemini Robotics-ER 1.5 is being made available to developers today via the Gemini API in Google AI Studio. Gemini Robotics 1.5 is currently available only to specific partners.
Source: Gemini Robotics 1.5 brings AI agents into the physical world（HN 69pt・12コメント）（HN Search (backfill)、2025-09-26）