AI edge platform for rapid prototyping

02 December 2021, 20:35

Cloud computing makes it possible to easily access and deploy servers on the Internet. Many of today’s popular Internet applications are based on cloud technologies and deployed on data centers across the globe. Over the years, these data centers have become the core backbone infrastructure of everything we do on the Internet. In a way, they have become the very foundation of modern computing. Whether you type a chat message, listen to Spotify or watch a Netflix movie, all network traffic is eventually routed to a datacenter somewhere where all the processing take place. Smart phones are simply our gateways to the datacenters.

While this centralized model works well for many of today’s applications, it may not be well suited for tomorrow’s applications such as computationally-intensive AI applications or simulated virtual world environments requiring significantly higher bandwidth and much lower network latency than traditional cloud computing services could ever provide. Hence, there is a need for a shift towards a more decentralized service delivery model.

Edge computing makes it possible to place computations at the “edge” of the network closer to where the data is generated, thus eliminating the need of faraway datacenters. The promise of 5G networks combined with edge computing has the potential to enable development of completely new applications and innovations that can change our world and the way we interact with the Internet.

RISE and FieldRobotix are currently developing an AI Edge platform together that makes it easier to develop and deploy computer vision applications on edge infrastructures. The platform automates deployment and provides an easy-to-use API for media processing and inference, thus taking away many barriers for rapid prototyping of new AI applications.

The video below shows an example of what could be developed in the future using this AI Edge platform and 5G. Cyborg assassins are of course not what we are looking for, but should rather be seen as entertainment. Who doesn’t like the Terminator movies anyway? However, it is not too difficult to imagine the benefits of being able to send in a drone into a razed mine to look for survivors. More importantly, being able to offload the drone could enable many interesting use cases.

Today, drone data must be processed offline, and it is also a hassle just to copy the data to a server for processing. To produce the video, several videos were recorded in Mjölkuddsberget in Luleå and then fed into the AI Edge platform to perform object tracking using the SORT algorithm. In a near future, we are aiming tobuild a complete system using 5G technologies in a real mine so that data can be analyzed in real-time while flying. We are also aiming at developing more advanced AI models to solve real-world problems while also investigate implications on battery consumption and demonstrate the benefits of edge computing.

Link to video: http://194.28.123.162:8000/MjolkuddsbergetFinal.mp4

Johan Kristiansson

Senior Forskare

+46 10 228 42 58

Nuria Agues Paszkowsky

Forsknings- och utvecklingsingenjör

+46 10 228 45 19