Gemini Robotics-ER 1.6: Powering real-world robotics tasks through enhanced embodied reasoning

Source: DeepMind Date: 2026-04-13 URL: https://deepmind.google/blog/gemini-robotics-er-1-6/

Summary

Google DeepMind released Gemini Robotics-ER 1.6, a reasoning-first embodied AI model with enhanced spatial understanding, multi-view success detection, and a new instrument-reading capability (gauges, sight glasses, digital readouts). Instrument reading: 86% success with ER 1.6, 93% with agentic vision (visual reasoning + code execution), versus 23% for ER 1.5. Substantially outperforms ER 1.5 and Gemini 3.0 Flash on counting, spatial reasoning, and ASIMOV safety benchmarks. Developed with Boston Dynamics.

Implications

23% → 93% on instrument reading is the industrial robotics unlock. Gauge reading, pressure monitoring, and sight glass inspection are core factory inspection tasks that currently require human technicians. ER 1.6 at 93% with agentic vision is approaching production deployment quality for industrial inspection automation. Boston Dynamics involvement signals this is being validated in real industrial environments.

Agentic vision (reasoning + code execution) as a robotics capability. The 7-point improvement from 86% to 93% by adding code execution to visual reasoning is a clean demonstration that the agentic pattern — not just vision, but vision + tool use — is more capable than pure vision for structured interpretation tasks. This is the robotics equivalent of LLM tool-use gains.

ER 1.6 vs. ER 1.5 quadrupling performance on instruments validates the version cadence. The ER line (1.5 → 1.6) is improving rapidly on specialized embodied reasoning tasks. The rate of improvement suggests Google is actively tuning for industrial robotics use cases, not just research demonstrations.

Watch:

• Boston Dynamics commercial product integration timeline for ER 1.6 capabilities
• Whether the instrument-reading capability generalizes to digital interfaces and screens (not just analog gauges)
• ASIMOV benchmark scores for ER 1.6 vs. 1.5 — safety performance during task execution is the deployment gate