September 2025 – Google has taken a significant step toward advancing artificial intelligence in the physical world with the launch of Gemini Robotics 1.5 and Gemini Robotics-ER 1.5, two powerful AI models designed to bring agentic reasoning, planning, and action into robotics. The announcement highlights a pivotal move toward general-purpose robots capable of perceiving, thinking, and acting with human-like reasoning.
Bringing AI Agents Into the Physical World
Robots have long struggled with tasks requiring contextual understanding and multi-step reasoning. Simple commands often demand complex problem-solving, like looking up local recycling rules, identifying objects, and executing precise physical actions. To address these challenges, Google has introduced a dual-model agentic framework that allows robots to not only act but also think before acting.
- Gemini Robotics 1.5: A vision-language-action (VLA) model that translates visual inputs and instructions into motor commands. Unlike traditional models, it generates a reasoning process before acting, making robotic decision-making more transparent and reliable. It also shows the ability to learn across different robot forms, accelerating skill development.
- Gemini Robotics-ER 1.5: A vision-language model (VLM) optimized for embodied reasoning, effectively serving as a robot’s high-level brain. It excels at spatial understanding, planning, and logical decision-making, while also leveraging digital tools like Google Search to enhance problem-solving.
Together, the two models enable robots to orchestrate complex, multi-step tasks, improving generalization across diverse environments and tasks.
Real-World Applications and Capabilities
The power of Gemini Robotics 1.5 lies in its multi-level thinking process. For instance, when sorting laundry by color, the model breaks down the task into logical steps: understanding the rule, identifying each clothing item, planning the movement, and executing precise motor actions.
Meanwhile, Gemini Robotics-ER 1.5 demonstrated state-of-the-art performance across 15 academic embodied reasoning benchmarks, including ERQA, Point-Bench, RoboSpatial-VQA, and VSI-Bench. Its ability to interact in natural language, predict trajectories, detect task progress, and call third-party functions makes it one of the most advanced AI reasoning systems for robotics.
Another breakthrough feature is cross-embodiment learning. Skills learned by one robot, such as ALOHA 2, can be transferred seamlessly to others like Apollo (Apptronik’s humanoid robot) or Franka’s bi-arm system, significantly reducing training time and broadening real-world usability.
Prioritizing Safety and Responsible AI
As AI takes on more physical roles, safety remains a top concern. Google has built robust safeguards into Gemini Robotics models.
- High-level semantic reasoning allows robots to evaluate safety before acting.
- On-board safety subsystems ensure low-level protections like collision avoidance.
- Alignment with Gemini Safety Policies enables respectful and reliable human interaction.
Google also introduced an upgraded ASIMOV benchmark, designed to test and improve semantic safety, including new datasets, safety-focused scenarios, and video modalities. In evaluations, Gemini Robotics-ER 1.5 showed best-in-class performance, demonstrating its ability to operate responsibly in dynamic human environments.
A Step Toward AGI in the Physical World
The launch of Gemini Robotics 1.5 marks a milestone toward Artificial General Intelligence (AGI) in robotics. By combining perception, reasoning, and tool usage, these models move beyond reactive machines, creating agentic robots capable of adapting and generalizing to new challenges.
Developers can now access Gemini Robotics-ER 1.5 via the Gemini API in Google AI Studio, while Gemini Robotics 1.5 is currently available to select partners. Google emphasizes that this innovation is just the beginning of a new era of intelligent physical agents, with enormous potential to reshape industries ranging from healthcare and logistics to manufacturing and household assistance.
Outlook
With Gemini Robotics 1.5, Google has laid the foundation for a future where robots are not just tools, but collaborative partners capable of navigating the complexities of the real world. By merging cutting-edge AI reasoning with physical action, these advancements bring us closer to building robots that are adaptable, intelligent, and safe for everyday environments.
As Google continues to expand its research and partnerships, the robotics community eagerly anticipates the next wave of breakthroughs that could redefine human-robot interaction.
Read more: Marina Ammirato to Head Landor New York as Managing Director, Driving Innovation and Growth
