Google DeepMind has officially launched Gemini Robotics-ER 1.6, a specialized vision-language model (VLM) designed to serve as the “brain” for next-generation autonomous robots. Announced yesterday, April 14, the model is now available in preview through the Gemini API and Google AI Studio.
The “ER” stands for Embodied Reasoning, signaling Google’s pivot from digital-only AI to models that can safely and intelligently navigate the physical world.
1. Key Upgrades: Seeing Like a Human
Gemini Robotics-ER 1.6 introduces “multi-view reasoning,” allowing robots to integrate data from several cameras simultaneously to understand 3D space.
- Instrument Reading: In a major leap for industrial automation, the model can now read analog pressure gauges, thermometers, and digital displays. It uses Agentic Vision (a mix of visual reasoning and code execution) to “zoom in” on a dial and interpret the value.
- Point-and-Reason: The model’s pointing accuracy has improved significantly. It can now handle complex spatial requests like: “Point to all objects that could fit inside this blue cup.”
- Multi-View Success Detection: Robots can now verify if a task is actually finished (e.g., “Is the pen fully in the holder?”) even if their view is partially blocked or the lighting is poor.
2. The Safety “Hard-Block”
Given that these models control heavy machinery, Google has implemented its most rigorous physical safety framework to date.
- Constraint Compliance: The model has a 10% improvement in recognizing potential injury risks in video scenarios compared to Gemini 3.0 Flash.
- Adversarial Safety: It is designed to reject “harmful” spatial commands that could cause a robot to collide with a human or damage delicate equipment.
3. Performance Metrics (vs. ER 1.5)
The internal benchmarks released by DeepMind show a dramatic jump in “success rates” for real-world physical tasks.
| Task Category | Success Rate (ER 1.6) | Improvement |
| Pointing & Counting | 80% | Significant leap |
| Single-View Detection | 90% | Most reliable |
| Multi-View Detection | 84% | New capability |
| Safety Compliance | Top Tier | +6-10% vs. Flash 3 |
4. Availability & Deprecation
Google is moving fast to migrate its robotics partners to the new architecture:
- Current Access:
gemini-robotics-er-1.6-previewis available now for developers. - Shutdown Notice: The previous model, Gemini Robotics-ER 1.5, will be officially shut down on April 30, 2026.
5. Why It Matters for You
As someone tracking the 27 million developer surge in India and TCS’s AI revenue, this model represents the “industrialization” of AI:
- Digital-to-Physical: This isn’t just about chatbots; it’s about software that can operate an oil rig, a warehouse, or a hospital.
- The “Agentic” Shift: Because the model can natively invoke Google Search and custom functions to solve problems it doesn’t understand, it reduces the need for constant human supervision.
