# Grabette: Robot Manipulation Data Recording > Source: > Published: 2026-07-23 19:03:34+00:00 # Grabette: Robot Manipulation Data Recording The core value here is the "open" nature of the system. Instead of being locked into a proprietary hardware-software stack, Grabette allows for a more flexible AI workflow when gathering the real-world data needed to teach robots how to interact with objects. It essentially bridges the gap between raw robotic movement and the structured data formats required for modern machine learning. ## Getting Started with Grabette For those looking for a practical tutorial on implementing this in their lab or home setup: 1. **Hardware Integration:** Connect your robotic arm and sensors to the Grabette recording interface. 2. **Configuration:** Define your data streams (joint positions, end-effector coordinates, and camera feeds) in the config file. 3. **Data Capture:** Execute the recording script while manually guiding the robot or running a teleoperation sequence. 4. **Export:** Save the captured trajectories into a format compatible with your training pipeline. ``` # Example: Starting a recording session (hypothetical command) grabette record --device arm_01 --duration 60s --output ./dataset/trial_01 ``` Whether you are a researcher or a hobbyist, this is a fantastic way to build a custom dataset from scratch. It removes a lot of the manual overhead, making the path from "physical movement" to "trainable data" much smoother. It's a great example of how open-source tooling can accelerate the development of embodied AI. [Next AI Guardrails vs. Offensive Security Research →](/en/threads/2483/)