FlowAvatar: Real-Time Full-Body Avatars from Sparse Egocentric Inputs on Consumer XR Devices

In plain English

A photoreal full-body avatar driven entirely by what a consumer headset already senses: head and hand tracking for body pose, plus the headset's own hand, face and eye tracking, all mapped onto one SMPL-X mesh, with a one-time enrolment from a short phone video for texture and body shape. Reported results are 5.40 cm MPJPE (roughly level with the HMDPoser baseline) with markedly lower jitter, 247 FPS streamed to an RTX 4090 laptop, and 27 FPS with 97 ms motion delay running entirely on a Quest Pro. The on-device figure is the notable one — photoreal full-body telepresence without a tethered workstation — though the two user studies (N=12 and N=10) are small and the authors call them preliminary.