The Safari SDK provides a comprehensive toolkit for the full lifecycle management of Google DeepMind's Gemini Robotics models. It targets researchers and developers working with robotic systems, enabling them to access, serve, evaluate, fine-tune, and manage model checkpoints and data.

How It Works

The SDK offers a Python-based interface and a command-line tool (flywheel-cli) for interacting with the Gemini Robotics platform. Key functionalities include model serving, data ingestion, model training, and artifact management. This integrated approach streamlines the workflow for developing and deploying advanced robotics models.

Quick Start & Requirements

• Install via pip: pip install safari_sdk
• Source code available on GitHub.
• Building wheel: scripts/build_wheel.sh
• Access to Gemini Robotics models and features requires joining the Gemini Robotics Trusted Tester Program.

Highlighted Details

• Supports the Gemini Robotics model series, including on-device models.
• Libraries for data logging (safari/logging), model inference (safari/model), and checkpoint management/data upload/fine-tuning (safari/flywheel).
• Includes examples for robot and simulation evaluation, with Aloha-specific code.
• flywheel-cli for training, serving, data management, and artifact download/upload.

Maintenance & Community

The codebase is under active development. User guides are updated for Trusted Testers.

Licensing & Compatibility

The README does not specify a license. Compatibility for commercial or closed-source use is not detailed.

Limitations & Caveats

Full functionality is contingent on participation in the Gemini Robotics Trusted Tester Program. The codebase is in active development, implying potential for breaking changes and evolving APIs.