A curated reference collection of technical papers, benchmarks, datasets, physics simulators, and developer SDKs for Google DeepMind's Gemini Robotics 2.0, Gemini Robotics ER 2, and VLA 2.0 physical AI ecosystem.
| Milestone / Publication | Technical Summary | Link |
|---|---|---|
| Gemini Robotics 2 (July 2026) | Whole-body intelligence (feet to fingertips), dexterous manipulation, and multi-robot collaboration. | DeepMind Announcement |
| Gemini Robotics ER 2 | Embodied Reasoning (ER) models specialized in 3D metric spatial grounding, dynamic task planning, and continuous multimodal video reasoning. | Google AI Studio ER Hub |
| ASIMOV-Agentic Benchmark | Safety evaluation benchmark measuring agentic physical AI safety refusal, velocity compliance, and human proximity handling. | Safety Benchmark Paper |
| Open X-Embodiment 2 | The expanded cross-embodiment multi-robot dataset powering generalist robot policies across humanoids, quadrupeds, and manipulators. | Project Page |
| RT-2: Vision-Language-Action | Foundational VLA work translating internet-scale vision and language into physical control tokens. | DeepMind RT-2 Blog |
| PaLM-E | Ancestral embodied multimodal language model bridging sensor signals with language planners. | ArXiv Paper |
- Google GenAI SDK (
google-genai): Official modern Python SDK with native typing, Structured Outputs, and Files API. - Google AI Studio: Fast interactive web environment for prototyping Gemini Robotics ER prompts and spatial grounding.
- Gemini API Cookbook: Official repository of code recipes and multimodal patterns.
- ROS 2 Gemini Bridge: Drop-in ROS 2 package for image subscription and task planning.
- NVIDIA Isaac Sim / Isaac Lab: GPU-accelerated photorealistic simulation standard for Sim-to-Real transfer.
- MuJoCo: Open-source physics engine maintained by Google DeepMind for contact-rich multi-finger manipulation.
- Gazebo Harmonic / Ionic: Standard robotics simulator tightly integrated with ROS 2.
- Open-ENPIRE Hand: Open-source compliant anthropomorphic gripper with tendon actuation.
- Universal Robots (UR5e / UR10e): Industry cobot arms.
- Franka Emika Panda / FR3: Research standard torque-controlled manipulator.
- Apptronik Apollo / Boston Dynamics Atlas: Humanoid platforms partnering with Gemini Robotics 2.
- Google DeepMind: Physical AI & The Future of Robotics: Technical overview of whole-body intelligence and cross-embodiment transfer.
- Google I/O: Multimodal Reasoning with Gemini: Understanding long-context video comprehension and spatial tokens.
Maintained by Pruthvi Geedh • Gemini Robotics 2.0 Community