Applying for Meta's Project Aria Research Kit

Today we finished the draft of our application for Meta's Project Aria Research Kit the sensor-packed research glasses we hope will become the front end of our entire interaction stack. Our supervisor will be submitting it on our behalf. Our project is about building a generalizable egocentric interface layer for robot control: one interaction stack that transfers across robot morphologies instead of being rebuilt for every platform. We validate this on two deliberately different robots we've built ourselves Heracles, our quadruped robot dog. The core research question: can gaze, gesture and speech, captured from the operator's own point of view, be mapped into a shared spatial representation that any robot can consume? Aria is essentially the only wearable, hands-free device that exposes research-grade perception rather than hiding it behind a rendering pipeline. The functions we've built our architecture around are the SLAM cameras and 6DoF trajectory (placing operator and robot in one shared coordinate frame without it, "put that over there" has no ground truth), eye tracking as a pointing modality, hand tracking for gesture input, and the microphone array for spoken commands. Fused together, these enable genuine multimodal interaction: you look at an object, gesture, say "put that there" and the system resolves all three references into one robot action. In the application we tried to make two things crystal clear. First, our architecture has two interaction modes resting on the same shared coordinate frame: direct teleoperation, where continuous 6DoF operator pose drives robot motion, and high-level symbolic commands the robot plans against. Second, the robot is the display. Aria has no screen, and that's a deliberate design choice on our end feedback to the operator is the robot's own motion in shared physical space. Our research question is precisely whether egocentric input alone can carry the interaction. While the application makes its way to Meta, work continues on the robot side: low-level actuator control on Heracles, the ROS bridge that will eventually consume Aria's streamed data, and a locally hosted LLM pipeline for parsing spoken instructions into structured commands. If the application goes through, the glasses slot straight into an architecture that's already taking shape. Fingers crossed. 🤞