Issue 40 Unmanned Systems Technology October/November 2021 ANYbotics ANYmal D l AI systems focus l Aquatic Drones Phoenix 5 l Space vehicles insight l Sky Eye Rapier X-25 l FlyingBasket FB3 l GCS focus l AUVSI Xponential 2021

42 Focus | AI systems one processor to another if the neural network relies on a specialist accelerator. It may be though that an AI framework can be moved to a number of general- purpose cores in an emergency to provide core functions. AI in space The AI for air traffic control also lends itself to controlling swarms of robots in space applications. The Human and Explainable Autonomous Robotic System (HEART) algorithm for example will train robots to work together on mining, excavation and even building tasks, and allow the robots to improve their collaboration skills over time. The robots will be trained on Earth but used as a fully autonomous swarm that doesn’t need to receive instructions from Earth to mine materials and construct simple structures. Exploring small bodies such as asteroids for mining is a challenge owing to the low-gravity fields and uncertainty in the gravitational environment. Attempting surface missions with inadequate gravity field information brings a high risk of failure. This can be mitigated by using swarms of craft, some orbiting the asteroid to measure the environmental conditions and feeding that data to the mining craft on the surface. This is a similar AI challenge to the air traffic control system. Conclusion Machine learning is becoming ubiquitous in many areas of unmanned systems. The technology started out as large, monolithic arrays for processing video images, but it has evolved into a series of pipelined cores that provide more performance for less power consumption, allowing more detailed analysis of images for safer system operation. Machine learning algorithms are also being used for signal conditioning, to improve the quality of data from navigation sensors. This can improve the accuracy of an unmanned system’s position and monitor the reliability of the sensor. Machine learning is also capable of controlling thousands of aircraft in the sky, keeping them separated regardless of the conditions. This capability will be key to controlling swarms of autonomous craft. However, all these systems have to be kept safe and operating safely. Machine learning can monitor the activity of a system to protect against security breaches and the failure of hardware or software modules. New software frameworks are being explored to allow modules to be swapped around in the event of a security breach or module failure, whether in software or hardware. Acknowledgements The author would like to thank Xavier Orr at Advanced Navigation, Tony King-Smith at AImotive, Pulin Desai at Cadence Design Systems Tensilica, Shai Kurianski at Airwayz, Amol Parikh at Doodle Labs and Will Keegan at Lynx for their help with researching this article. October/November 2021 | Unmanned Systems Technology Machine learning can identify suspicious activity in an unmanned system to provide greater security (Courtesy of Secure CAV) Using machine learning to monitor the system performance and replace failing units requires a modular software structure (Courtesy of Lynx Software Technologies)