Can real-time artificial intelligence techniques be applied to synthetic environments?
That’s the question occupying some of the best and brightest minds in the industry. If it can, we will see a step change in how synthetic environments are developed, applied and operated. One of Thales‘s senior design engineers, Graham Long, discusses some of the challenges ahead and looks at the potential rewards of this innovative approach.
Synthetic environments play an essential role in a wide range of immersive training, battle simulation, mission planning, product development, factory design and many other applications. Data, of course, is key. All else being equal, the greater the data the better the representation of reality. Happily, there is no shortage of data.
Fleets of inexpensive mini-satellites now capture images of the entire earth every day. Drones harvest untold volumes of data on-demand. Social media ‘data lakes’, and an estimated 20 billion Internet of Things (IoT) devices, add to the glut of available data in all of its various forms.
The problem lies not in the volume or availability of data, but in how to incorporate it in a synthetic environment - and how to keep that environment refreshed in real-time as new data floods in.
At the time of writing, it takes weeks if not months to create a synthetic environment. The stark reality is that, as long as there is a disparity between the time it takes to capture the data and the time it takes to apply it, real-time synthetic environments will remain a dream.
There are ways to mitigate the lag between capturing the data and applying it, but they all have drawbacks and none of them get near to solving the real problem, which is to provide an, end-to-end solution that can not only handle massive data inputs but seamlessly process it and serve it to a synthetic environment.
Learning the lessons
Outside of the synthetic environment sector, enterprise is already processing and exploiting huge amounts of data. The proliferation of IoT devices , the rise of drones, the promise of autonomous vehicles, the science of face and speech recognition, object detection and full-motion video analysis - all of these have accelerated the demand for AI and informed its development.
Closer to home, and of particular interest, artificial intelligence is enabling the integration of satellite and airborne sensors to rapidly create digital maps. These illustrate the state of roads, buildings, forests, waterways and other features on a truly impressive scale. One recent project mapped all of the 170 million buildings in the United States.
Are there lessons to be learned here? Could artificial intelligence be harnessed to solve the data glut problems of synthetic environments? With additional research and experimentation with Artificial Intelligence, Neural Networks, Deep Learning, Inference and all of the various subsets of these and their related technologies and processes, then yes, I believe it can.
I see no reason why, with the proper application of artificial intelligence, the cumbersome offline data preparation and processing tasks, common to synthetic environments today, can’t be encapsulated into a pipeline that can intelligently analyse, process and exploit input data; extract features, and generate content on the fly, in a seamless end-to-end, process.
And when we do, we’ll have taken the very first steps on a journey to a time when synthetic environments are effectively indistinguishable to real life.
You can download a full version of the paper from the Documents section below.
About Graham
Graham Long has worked in the simulation industry for over 30 years, starting as a synthetic environment developer and progressing to directing the development of synthetic environment generation and authoring tools. Widely recognised as one of the industry’s leading experts, he has been a major contributor to more than 20 civil and military flight simulation projects. Today, he continues to push back the boundaries of synthetic environment development and generation technologies.