Gemini Robotics

Architecture

VLA built on Gemini 2.0 with physical actions as output modality, plus an intermediate reasoning layer for spatial analysis and safety. Gemini Robotics-ER is the embodied-reasoning VLM companion. Gemini Robotics On-Device runs locally, network-independent, and is fine-tunable with 50-100 demonstrations. Gemini Robotics 1.5 emphasizes a 'think before acting' chain-of-thought style.

Key facts

DEPLOY maturity rationale

Most advanced DeepMind VLA; commercializing via Apptronik; not at-scale production. Re-derived as research-to-commercial; recorded here as research given current scale

Announced

March 2025

Apptronik partnership

Confirmed; Gemini Robotics powers Apptronik Apollo in joint roadmap

On-Device variant

First DeepMind VLA available for fine-tuning; 50-100 demonstrations

Cross-embodiment

Trusted-tester robots; cross-platform deployment

Apollo partnership type

The Apollo (Apptronik) relationship is investment-linked: Google led Apptronik's Series A and the Gemini Robotics partnership formalizes the technology integration. Distinct from Atlas-Gemini, which is a dedicated bidirectional co-development with Boston Dynamics. Apollo also operates inside NVIDIA's Isaac ecosystem; primary-source evidence for GR00T as Apollo's onboard brain at runtime is not yet established, so DEPLOY does not record a GR00T powers relation on Apollo (only the Gemini Robotics powers relation).

Developed by (1)

• Google DeepMind

Powers (1)

• ApolloApptronik · humanoid

Sources (2)

1. Google DeepMind: Gemini Robotics (primary) · https://deepmind.google/models/gemini-robotics ↗
2. Google DeepMind Blog · https://deepmind.google/blog/ ↗