Google DeepMind unveils Gemini Robotics 2.5 for dexterous manipulation and factory kitting

Rolling out to trusted testers across Apptronik’s five-fingered Apollo 2 humanoid and Franka Duo arms, Gemini Robotics 2.5 lifts manipulation success by up to 100% and hits 96% autonomous kitting accuracy in Mercedes-Benz factory pilots.

Google DeepMind unveiled Gemini Robotics 2.5 for trusted testers and robotics partners on 10 November 2026, a day after NVIDIA introduced GR00T 2. Evaluated across more than 200 distinct tasks on platforms including Apptronik’s Apollo 2 humanoid (equipped with five-finger, 22-degree-of-freedom SharpaWave dexterous hands) and Franka Duo parallel-jaw arms, the vision-language-action model handles fine manipulation such as tying gift-box ribbons, operating kitchen tongs, and pipetting liquids.

Bar chart showing Gemini Robotics 2.5 outperforming Gemini Robotics 2 across tabletop pick-and-place tasks on Apollo 2 and Franka Duo — Tabletop pick-and-place task success rates comparing Gemini Robotics 2.5 with Gemini Robotics 2.Image: Google DeepMind

Mercedes-Benz factory pilots and ASIMOV safety

In live automotive manufacturing pilots with Apptronik and Mercedes-Benz in Kecskemét, Hungary, and Sindelfingen, Germany, engineers paired Gemini Robotics-ER for high-level spatial planning with Gemini Robotics 2.5 for low-level dexterous control, reaching a 96% autonomous kitting success rate on Apollo 2. DeepMind also added real-time inference-time safety guidance that steers arms clear of moving obstacles and nearby workers, alongside open-sourcing the ASIMOV-Jailbreak physical safety benchmark.

Bar chart showing Gemini Robotics 2.5 improving multi-step and contact-rich manipulation success by 30% to 100% over Gemini Robotics 2 — Multi-step and contact-rich manipulation results across dexterous hand and dual-arm platforms.Image: Google DeepMind

Key numbers

96% autonomous automotive kitting success rate on Apollo 2 in Mercedes-Benz factory pilots
20% to 30% faster task completion across more than 200 evaluated physical manipulation tasks
33% higher safe completion around physical obstacles and 43% higher around nearby people
99.5% adversarial prompt resistance on ASIMOV-Jailbreak when paired with Gemini 4 Argon (up from 82% on Gemini Robotics-ER 2)

Key details

Access status

Trusted tester and partner rollout announced on 10 November 2026

Evaluated robot hardware

Apptronik Apollo 2 humanoid (five-finger, 22-degree-of-freedom SharpaWave hands) and Franka Duo dual arms

Hierarchical control split

Gemini Robotics-ER for spatial planning paired with Gemini Robotics 2.5 for dexterous control

Factory pilot sites

Mercedes-Benz plants in Kecskemét, Hungary and Sindelfingen, Germany (96% kitting accuracy)

Open safety benchmark

ASIMOV-Jailbreak physical safety suite (99.5% adversarial resistance with Gemini 4 Argon)

Release details & context

Synthetic digital-twin practice, contact-rich gains and industrial partners

In Google DeepMind’s Gemini Robotics 2.5 post, DeepMind details collaborations with Agile, Apptronik and Mercedes-Benz across more than 200 physical manipulation tasks, improving multi-step and contact-rich task completion by 30% to 100% over Gemini Robotics 2 across tasks ranging from laundry folding and pastry service to picking metal automotive brackets from supply bins into sequencing trays.

To reduce real-world trial-and-error wear alongside world-simulation work like Google DeepMind’s Genie 4 and NVIDIA’s GR00T 2, engineers used Gemini 4 Argon to construct digital-twin simulation scenes from camera views, generate thousands of randomised practice episodes and diagnose failed robot videos automatically.

Common questions

How did Mercedes-Benz and Apptronik structure the Gemini Robotics 2.5 factory kitting stack?

At Mercedes-Benz plants in Kecskemét and Sindelfingen, Apollo 2 humanoids with 22-degree-of-freedom SharpaWave hands paired Gemini Robotics-ER high-level spatial planning with Gemini Robotics 2.5 motor control to achieve 96% autonomous kitting success.

What physical safety benchmark did Google DeepMind open-source alongside Gemini Robotics 2.5?

DeepMind open-sourced ASIMOV-Jailbreak, where pairing Gemini Robotics 2.5 with Gemini 4 Argon blocked 99.5% of adversarial physical-safety prompts, up from 82% on Gemini Robotics-ER 2.

Google DeepMind release timeline (8 launches tracked)

Earlier27 Oct 2026

Google DeepMind unveils Genie 4 real-time interactive world model

Google DeepMind released Genie 4 as a preview to trusted testers, bringing interactive world generation to the Gemini API and Project Genie.

Earlier28 Oct 2026

Google DeepMind unveils Gemini 4 Carbon frontier model

Google DeepMind’s Gemini 4 Carbon scores 59 on Artificial Analysis v5.0 and 88.2% on SEC-bench Pro at Argon’s $2/$10 per million token API price.

Earlier29 Oct 2026

Google DeepMind publishes economic framework for multi-agent AGI regimes

Google DeepMind modelled four AI capability regimes, finding that multi-agent coordination thresholds trigger abrupt jumps in firm concentration and wages.

Shipped next11 Nov 2026

Google DeepMind releases AlphaGenome 2 API for single-cell DNA variant prediction

Google DeepMind launches AlphaGenome 2 with custom scRNA-seq conditioning, imputed tracks and CpG methylation scoring across 1 Mb DNA windows.