Page 382 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

Use Case 19: Bandwidth efficient live interaction with virtual 3D

demonstrator using semantic communication and GenAI

Organization: Tata Consultancy Services

Country: India

Contact Person(s):
Abhijan Bhattacharyya, Abhijan.bhattacharyya@ tcs .com
Ashis Sau, ashis.sau@ tcs .com
Suraj Mahato, surajkumar.mahato@ tcs .com

1 Use Case Summary Table

Item Details

Category 5G
Problem
Addressed Clearly describe the primary issue or challenge
1) 3D remote education experience with very low bandwidth: It enables train-
ees or students to interact with a live 3D virtual representation of a distant
teacher/ demonstrator in real-time with a very low-bandwidth consumption.
The bandwidth savings is very significant compared with 3D and even
with 2D visual transmission which is in the order of several Gbps and
Mbps respectively.
2) Democratized scalable solution with no need for specialized infra-
structure: Unlike holographic 3D video transfer, it does not require
any specialized sensor, camera and studio set up at both end points.
The teacher/ demonstrator only needs a RGB camera attached to a
Key Aspects of computer. This solution can be in generalized as cost-effective, scalable
Solution 3D telepresence with acceptable realism.
3) AI-native Semantic Communication with optional GenAI integration: It
uses semantic communication to predict the body pose and position of
the distant teacher / demonstrator in real-time using AI and transmits
the predicted information to the trainee / student-end for mimicking a
pre-stored 3D body model. It uses artificial intelligence to predict the
body posture of the distant teacher in real-17th European Conference
on Technology Enhanced Learningtime, and the figure is displayed
in situ using augmented reality. GenAI may be used to virtually recre-
ate some of the objects associated with the teacher, at the trainee/
student’s end.

Technology Semantic Communication, human pose estimation, natural language
Keywords processing, augmented reality.

For body modelling we used the datasets used in [2]. For motion prediction
Data Availability
we used [3].

346