Page 384 - AI for Good Innovate for Impact
P. 384
AI for Good Innovate for Impact
net specialized set or h/w unlike the systems that require special body-sensors to be attached
to get the body pose [10]. The actual transmission over the network backhaul takes only about
a few hundred bytes to a few kilo bytes per frame. This is much less than 3D video streaming
or even 2D video. The solution may also be deployed for rendering on digital video screen if
AR/VR glass and associated infrastructure is not available.
As a further enhancement, GenAI service may be added to the receiving end to replicate
the necessary objects relevant to the remote trainer. As shown in Fig. 2, the remote trainer is
demonstrating a robot which does not exist at the trainee’s end (observer in the figure). The
transmit-side intelligence may generate and send a scene description along with the 3D pose
information. The GenAI service understands the objects and the relations from the scene
description and recreate those objects.
Use Case Status:
An initial PoC to understand the feasibility has been done and the relevant experiences have
been reported in the publication in [1].
348
