Meta has recently disclosed comprehensive technical details of its Aria Gen2 research glasses for the first time, providing the most detailed technical analysis of the device since its release in February this year. Compared to the first-generation product released in 2020, Aria Gen2 represents a comprehensive technological leap forward.
Comprehensive Upgrade in Hardware Design
The new generation of glasses weighs between 74 and 76 grams, offering eight different frame designs to suit various face shapes. The foldable temple design enhances portability, with battery life lasting 6-8 hours to meet daily usage needs.
The most prominent upgrade in Aria Gen2 is the increase in computer vision cameras from two to four. The newly equipped global shutter sensors can simultaneously expose each pixel, completely solving the motion distortion problem during rapid movement.
The dynamic range of HDR has been significantly increased from 70 decibels in the first generation to 120 decibels, enabling the device to clearly capture both extremely bright and dark areas in a scene simultaneously. The stereoscopic overlap between the main cameras has also been improved from 35 degrees to 80 degrees, significantly enhancing depth measurement accuracy.
Innovative Sensor Technology
Meta has added three key sensors to Aria Gen2. The calibrated ambient light sensor not only distinguishes different light sources but also features a UV mode to identify sunlight and artificial lighting, achieving intelligent adaptive adjustment of camera settings.
Most notably, the contact microphone built into the nose pad picks up user sounds through nasal vibrations. This technology, based on structural sound conduction rather than air conduction, can work normally even in noisy environments. Wind tunnel tests show that when traditional microphones fail entirely, contact microphones can still capture whispers.
Significant Enhancement in AI Processing Power
Aria Gen2 uses a dedicated coprocessor for device-side machine learning, with all machine perception algorithms running directly on the glasses. The visual-inertial odometry system fuses camera images with motion sensor data to achieve precise six-degree-of-freedom position tracking.
The device also features complete eye-tracking functionality, capable of detecting gaze points, calculating convergence points, tracking blinks, and measuring pupil size. The 3D hand-tracking function can reconstruct finger joint positions in real time, laying the foundation for future AR interactions.
Aria Gen2 uses SubGHz radio technology, achieving sub-millisecond multi-device synchronization accuracy. Meta stated that this hardware-based synchronization method is significantly more precise than the previous software-based solution.
Research Plan for 2025
Meta will open research applications for Aria Gen2 throughout the year, allowing interested researchers to register for access. The company plans to demonstrate interactive demonstrations at the CVPR 2025 Computer Vision Conference from June 11-15.
Since its launch in 2020, the Aria project has generated multiple computer vision and robotics training datasets, including Ego-Exo4D. Meta is using these first-person video data to train AI assistants with the goal of creating a general-purpose AI assistant capable of assisting with daily tasks, competing with Google and OpenAI in the AI assistant field.
