Claimed to be a world first, the STARTLE system works by emulating a mammal’s conditioned fear-response mechanism to help autonomous vehicles avoid danger.
In the natural world, an animal will execute a primary task — such as foraging for food — while being attuned to what is going on around it, taking action if it senses an anomaly such as a predator.
In this way, the animal can make best use of its stored energy and energy it takes on board through feeding but can take evasive action if required.
Similarly, STARTLE, which has been tested in ground vehicle simulations, uses a combination of artificial neural network and diagnostic expert systems to continually monitor and assess potential threats.
‘What we’re emulating is the mechanism that allows mammals to have focus of attention, concentrating on what their primary task is,’ said Mike Hook, principal consultant at Roke.
Making use of existing hardware, STARTLE processes information from multiple on-board sensors, cueing systems to assess and confirm potential threats.
‘You’ve got a continuous stream of data that is being processed at quite a high level, looking at possible threats that it has been trained to detect,’ said Hook. ‘If it sees something that matches that, it will seek to prove whether there is a threat or not. This is what drops the false alarm rate.’
Hook added that STARTLE will make a request for additional data to be collected and for extra processing to be undertaken, should it spot a threat.
In this way, STARTLE makes best use of on-board electrical systems and Paul Webb, business sector manager for unmanned systems at Roke, believes this could be of benefit to new designs of autonomous vehicles, such as UAVs.
‘Unmanned systems in general will need intelligent power management, because they will be carrying multiple sensors that you might not need to use all the time,’ he said. ‘Essentially, this capability will enable them to make an intelligent decision about which sensors should be used and when.’
Webb added that STARTLE could be applied to manned systems or for other completely orthogonal domains.