Crowd control

A military-funded collaboration is to study nature for single-handed control of UAV swarms.


A major US research collaboration will investigate how wolf packs and shoals of fish could teach engineers to control swarms of unmanned vehicles.


The $5m (£2.6m) US military-funded project will develop complex algorithms that mimic group behaviour in nature. The research could allow one person to co-ordinate thousands of surveillance UAVs, ‘smart dust’ sensors or driverless vehicles.


Today’s unmanned robots need around 10 operators each, according to Prof Vijay Kumar, leading the research at the University of Pennsylvania. The Department of Defence-funded five-year programme also includes the Universities of California, Yale and MIT.



‘With complex systems like the robots repairing the Hubble telescope, that number grows,’ he said. ‘We want to reverse that ratio and have a hundred to a thousand units in a swarm controlled by a few people.’



A surveillance UAV swarm, for example, would be more effective and less vulnerable than one expensive aircraft, but today’s technology cannot simply be scaled up — the huge amount of data would be unmanageable for a ground crew.



Analysis of group behaviour in nature could hold the key — from the feeding habits of small-scale insects like harvester ants, or plankton, up to the hunting behaviour of lions or killer whales.



A major problem for controlling a swarm is finding a way to dispense with a designated leader, which would disable the whole group if destroyed, so the team hopes to find a way to control the swarm collectively.



‘If the UAVs were undertaking surveillance, the swarm would figure out automatically who does what; who will go to the east side of a building, who to the west and who will relay the information,’ said Kumar. If the swarm needs to locate itself, each unit would only need a limited view of its position, but the swarm’s collective intelligence would, for example, yield a complete picture.



The team aims to process a wealth of animal behaviour data in the hope that patterns emerge. A fundamental understanding of complex natural systems is unnecessary, said Kumar. ‘It’s very difficult to develop models of nature that are robust. You tweak a parameter and the whole system changes,’ he said. ‘Pure data analysis could well be more reliable.’



The researchers plan to reveal their biology-based algorithms within the first 12 months, leading to a full-scale demonstration of 30 UAVs and up to 100 ground vehicles in three years.


A number of other teams are also working on the unmanned swarm challenge, including one from the Defence Science and Technology Organisation in Australia. Researchers at BAE Systems’ Advanced Technology Centre are also planning trials in the Australian Outback later this year, in which a fleet of research UAVs will co-operate with a ground robot on surveillance tasks.