Neural Networks Play Guess-the-Material to Reduce Latency in the "Tactile Internet"

Researchers turn to artificial intelligence to reduce the latency inherent in remote control over the "Tactile Internet."

An AI helps reduce latency of remote handling over the "Tactile Internet." (πŸ“·: Mondal et al.)

A team of researchers led by Elaine Wong from the University of Melbourne is to present a neural network system which aims to resolve one of the biggest hurdles facing the so-called "Tactile Internet": Latency.

The "Tactile Internet" describes a world where human-machine interaction, typically but not exclusively in virtual reality, includes a sense of touch β€” such as being able to feel the surface of a virtual or remote object. While a number of solutions have been proposed, and in some cases even commercialised, for bringing that sense of touch to the user, they all have one major drawback: A latency between the user touching the virtual object and feeling the feedback, which shatters the illusion in as little as a single millisecond.

"These response times impose a limit on how far apart humans and machines can be placed," Wong explains of her team's research. "Hence, solutions to decouple this distance from the network response time is critical to realising the Tactile Internet."

"To facilitate human-to-machine applications over long distance networks," notes lead author Sourav Mondal, "we rely on artificial intelligence to overcome the effects of long propagation latency." That artificial intelligence: A reinforcement learning algorithm which works to guess feedback required before it can be known, dubbed Event-based Haptic Sample Forecast β€” or EHASAF.

Using EHASAF, researchers had users interact with a virtual ball via a pair of gloves which include sensors and feedback capabilities. The ball could be made of four different virtual materials, but which material is used wouldn't be clear until it was touched β€” yet EHASAF, the researchers claim, was able to predict the correct response with a 97 percent accuracy.

"We think it is possible to improve prediction accuracy with a greater number of materials," Mondal adds. "However, more sophisticated artificial intelligence-based models are needed to achieve that. More and more sophisticated models with improved performance can be developed based on the fundamental idea of our proposed EHSAF module."

The team is to present its work at the Optical Fiber Communication Conference and Exhibition (OFC) this March at the San Diego Convention Centre. No word have been given yet on a route to commercialisation, but more information is available on the OSA website.

Gareth Halfacree
Freelance journalist, technical author, hacker, tinkerer, erstwhile sysadmin. For hire:
Related articles
Sponsored articles
Related articles