Skip to main content

Project Aria Tools

projectaria-tools is the official library for accessing and processing data recorded by Project Aria glasses. It provides a unified interface to load VRS files and access all associated sensor and machine perception data through high-level, easy-to-use APIs. The library supports efficient operations such as random access, timestamp-based queries, and synchronized multi-sensor data retrieval, with a consistent set of C++ and Python APIs seamlessly bridged via Pybind11.

Project Aria glasses record its data in VRS (Visual Recordings System) format, which is a general-purpose recording format from Meta. Properly interpreting Project Aria VRS requires detailed understanding of the device’s sensor configurations, calibration models, and on-device perception outputs. projectaria-tools abstracts these complexities, ensuring that all data is correctly parsed, aligned, and represented for research and development workflows.

In addition to data access, the library includes visualization and conversion utilities that allow users to inspect and export selected streams to standard formats such as CSV and MP4, making it easier to analyze and share Aria data across diverse toolchains.

Getting Started

If you're new to projectaria-tools, start here:

Python Notebook Tutorials

The tutorials are runnable Jupyter notebooks kept in the GitHub repository rather than duplicated here, so what you read is always what you run. Open one in Colab to run it with no local setup -- each notebook installs projectaria-tools and downloads the sample data it needs.

TutorialRun itWhat it covers
1VrsDataProvider BasicsOpen in ColabLoad an Aria VRS file and access its streams.
2Device CalibrationOpen in ColabSensor intrinsics and extrinsics, projection, undistortion.
3Sequential Multi-Sensor AccessOpen in ColabThe queued API for streaming several sensors in timestamp order.
4Device Time AlignmentOpen in ColabTime domains, timestamp queries, and multi-device alignment over SubGHz.
5MPS BasicsOpen in ColabMPS output layout, MpsDataPathsProvider and MpsDataProvider.
6VIO and TrajectoryOpen in ColabDevice pose: on-device VIO, and the MPS trajectory and semi-dense point cloud.
7Hand TrackingOpen in ColabHands from both sources: the on-device stream and the MPS result.
8Eye TrackingOpen in ColabGaze from all three sources: on-device geometric, on-device ML, and MPS.
9Neural Band sEMGOpen in ColabThe Meta Neural Band emg stream, its batch model and its two clocks.

Tutorials 1-4 cover reading a recording. Tutorial 5 covers reading MPS output. Tutorials 6-8 each cover one algorithm across every source that produces it, on-device and MPS alike, since choosing between those sources is usually the real question. Tutorial 9 covers the Neural Band stream.

Visualization Tools

Tools for visualizing and exploring Aria data: