Google DeepMind
Google DeepMind's Robotics Research
Google DeepMind
Home > Companies > Google DeepMind
Overview
Google DeepMind is a pioneer in robotics AI, leading key advancements in VLA models from RT-1 to Gemini Robotics. The current organization was formed in 2023 through the merger of Google Brain and DeepMind.
| Item | Details |
|---|
| Headquarters | London, UK / Mountain View, CA |
| Founded | 2010 (DeepMind), 2023 (merger) |
| CEO | Demis Hassabis |
| Robotics Lead | Vincent Vanhoucke |
Key Models
Timeline
| Year | Model | Key Contribution |
|---|
| 2022 | RT-1 | First large-scale Robotics Transformer |
| 2023 | RT-2 | First VLA, “Action as Language” |
| 2023 | RT-X | Open X-Embodiment (33 research labs collaboration) |
| 2025 | Gemini Robotics | Gemini 2.0-based VLA |
RT-1 (2022)
- Trained on 130K episodes
- Generalized to 700+ tasks using Transformer architecture
- 97% success rate (on trained tasks)
RT-2 (2023)
- “First VLA” - Outputs actions as language
- Based on PaLM-E/PaLI-X (55B)
- Emergent capabilities: Understanding abstract commands
RT-X (2023)
- Collaboration of 33 research labs (Open X-Embodiment)
- 22 robot types, 160K+ episodes
- Open dataset release
Gemini Robotics (2025)
- Most powerful VLA based on Gemini 2.0
- 2x+ performance over SOTA
- On-Device version released (local execution)
Research Approach
Foundation Model Method
[Large-scale Pre-training] → [Fast Fine-tuning] → [New Task Adaptation]
Core Principles
- Scaling: More data, larger models
- Multimodal: Integration of Vision + Language + Action
- Cross-robot: Training/deployment across diverse robots
- Collaboration: Open research like Open X-Embodiment
Partnerships
| Partner | Details |
|---|
| Apptronik | Applying Gemini 2.0 to Apollo humanoid |
| Boston Dynamics | Testing Gemini Robotics-ER |
| Agility Robotics | Digit robot collaboration |
References
See Also