Unmanned Systems Technology 013 | AutonomouStuff Lincoln MKZ | AI systems | Unmanned Underwater Vehicles | Cosworth AG2 UAV twin | AceCore Neo | Maintenance | IDEX 2017 Show report

32 A rtificial intelligence (AI) covers a wide range of technologies in many diverse areas, from voice recognition, data mining and customer support services to driverless cars and unmanned aircraft control systems. One initial approach for autonomous vehicles centred on rule-based systems. The trouble with these though, especially for driverless cars, is that the range of all the possible options, called the problem space, becomes so large that it is almost impossible to describe a rule base that covers every permutation. As a result, machine learning systems are being developed to allow machines to learn how to come up with the knowledge to create the rules themselves, but that creates challenges for the safety and certification of a system. Machine learning covers several techniques, from deep learning algorithms that use neural networks to mimic how the brain recognises images, to pattern recognition. At a high level of abstraction, the networks are ‘trained’ using images that have been tagged with information, and they can then use this knowledge to accurately identify similar structures in new images or video feeds. However, AI for autonomous systems is not just about deep learning for image recognition; several other elements are needed for effective, safe AI systems. Overcoming every challenge of implementing artificial intelligence in unmanned systems is still some way off, but some solutions are beginning to emerge. Nick Flaherty reports Frames of mind April/May 2017 | Unmanned Systems Technology