Unmanned Systems Technology 003 | UAV Solutions Talon 120 | Cable harnesses | Austro Engine AE50R and AE300 | Autonomous mining | AUVSI 2015 show report | Transponders | Space systems

15 data – just like a pilot’s intuition. A computer can’t do that. A fly has a very complex flight system but it is controlled by only a handful of neurons, and we are pushing the memristor in that direction.” With synthetic neurons using memristor- based circuits, an external stimulus such as an image from a database causes the system to fire, and if it sees repeated stimuli that are similar then the rate of firing increases, so a reinforcement of the image takes place. At a physical level, the memristor’s resistance is a function of all the previous stimuli as well as the latest one – and that’s simply a voltage. That is how the ‘learning’ takes place. This is an advantage because if you had a conventional control system and you wanted to recognise objects in the environment then you’d need a database with image capture, but it’s impossible to have every image of a given object and access it in real time, as you have to keep scanning a very large database. However, the Bio Inspired neural net databases are much smaller by using a technique known as weighting. With this technique the database contains a representative tree for example, but it’s not exact, and the neural net gives a weighted probability of a match between a real-world image and the database version. That allows the recognition to be done in real time. Then there’s the challenge of context sensitivity, which changes the way the control system works according to the environment. Theoretically, if it’s a rural environment then you could reduce the amount of data that needs to be recognised, as there are fewer buildings, but this is not possible in an urban environment. Gafron says windows are a great example of how context sensitivity changes the way control systems operate. For a UAV flying through a window, a conventional system needs a range of proximity sensors, but with neural nets it identifies a window, checks whether it is open by comparing it to a database image of an open window, and if it can flies through it. The result is that you Terry Gafron | In conversation It’s quite possible for us to design a UAV-specific control chip whose sole purpose is autonomous control with all flight systems Unmanned Systems Technology | Summer 2015 Gafron has been chief technology officer of Bio Inspired Technologies since 2010, setting up the company after a spell as the research manager for advanced memories and reconfigurable technologies at Boise State University, Idaho. He has also been CEO since 2013, and has previously worked at memory chip maker Micron as well as at a solar cell manufacturer. The company is currently working on a Department of Defense project to develop an adaptive computing architecture using memristors to produce the world’s first modular intelligent processor that learns on its own. It has also developed a neural network processor called Cognimem for high-end video and data stream classification and identification. This is used for a portable adaptive intelligence and response system. Bio Inspired has also developed Prometheus, a fully autonomous GPS-guided personal UAV system enabling remote guidance and surveillance linked directly to the GPS satellite system, Google Earth, and the user’s local mobile phone provider. Terry Gafron A house fly has a complex flight system, and it’s in that direction that Bio Inspired is pushing its memristor development