$ cat wiki/models/gemini-robotics-er-1-6.md
Gemini Robotics ER 1.6
Spec
| Attribute | Value |
|---|---|
| Developer | Google DeepMind |
| Released | 2026-04-15 |
| Announced | 2026-04-15 |
| Context window | unknown |
| Pricing | unknown |
| License | unknown |
| Availability | Gemini API, Google AI Studio (developer access, as of 2026-04-15) |
| Full name | Gemini Robotics-ER 1.6 (Enhanced Embodied Reasoning) |
| Architecture | Gemini multimodal + robotics-specialized reasoning stack |
Release Date
Released April 15, 2026. Successor to Gemini Robotics-ER 1.5. Notable collaboration with Boston Dynamics on industrial inspection use cases.
Benchmarks
| Task | ER 1.6 | ER 1.5 | Notes |
|---|---|---|---|
| Gauge reading accuracy | 93% | 23% | Agentic vision pipeline |
| Multi-view 3D understanding | Improved (no number) | — | Spatial comprehension |
| Source: (DeepMind blog) |
The jump in gauge reading (23% → 93%) is the headline number — this represents a functional threshold crossing for industrial deployment.
Use Cases
- Industrial inspection (energy plants, manufacturing, chemical facilities)
- Gauge/meter reading for autonomous inspection robots
- Physical task planning and success verification
- Multi-step robotic manipulation
Capabilities
- Visual and spatial understanding: Multi-view scene comprehension, depth estimation
- Instrument reading: Reads analog gauges, pressure meters, sight glasses with 93% accuracy using agentic vision (observe → reason → verify)
- Task planning: Plans and sequences physical actions with success detection
- Safety: Described as "safest robotics model to date" (no specifics disclosed)
Compared To
| Model | Gauge Reading | Platform | Notes |
|---|---|---|---|
| Gemini Robotics ER 1.6 | 93% | Spot, robot arms | Google DeepMind |
| Gemini Robotics ER 1.5 | 23% | — | predecessor |
| RT-2 (DeepMind, 2023) | n/a | — | text-to-actions, earlier gen |
Boston Dynamics Collaboration
Capability development jointly with Boston Dynamics (Spot robot). Demonstrated in simulated industrial environments:
- Spot robot navigating plant floor
- ER 1.6 reading analog pressure gauges on pipes
- Agentic loop: move to optimal viewing angle → read gauge → verify → log
Significance
Instrument reading is one of the last practical barriers to autonomous industrial inspection. Manual inspection of gauges/meters is a major labor cost in energy, water, and manufacturing. ER 1.6 crossing 90% accuracy makes deployment viable for supervised automation.
See Embodied Agents for broader context on the physical AI trajectory.