Will the wars of the future be fought by intelligent, autonomous swarms of UASs able to reconfigure and adapt to any mission or situation? DANIEL PAGE, AMRAeS and Dr HORWARD TRIPP* assess the 2020 battlespace and their benefits and challenges.
This is a full article published in Aerospace International: July 2012
Populist perceptions of ‘robot’ swarms often conjure up images of hundreds of miniature machines working together in eerie harmony. An often cited example is the excellent work of the GRASP lab at the University of Pennsylvania(1) whose popular and impressive videos show several quad-rotor micro-UAVs (unmanned air vehicles) flying challenging manoeuvres in perfect unison.
Compared with more traditional single platform approaches, this ‘collective co-ordination’ of the swarm offers a number of attractive benefits. Valuable tasks such as reconnaissance and situational awareness could be greatly enhanced through the generation of multiple feeds or ‘takes’, integrated to form a dataset greater than the sum of its parts. Greater numbers of low-tech platforms could find fleeting targets far more quickly than a single, complex system and, crucially, the robustness to threats could be greatly increased.
A future swarm could be constructed in a number of ways, from a homogeneous set of platforms (e.g. a swarm of simple micro-UAVs or low-cost tactical systems) through to a layered set of different platforms, operating in ways that acknowledge their respective strengths and weaknesses. Such a concept could include a larger and more capable ‘mother’ platform (such as a MALE or tactical UAV) releasing a series of ‘child’ nano-UAVs towards a particular target or region of interest.
Building a swarm in an affordable manner could be tackled in a number of ways, from modifying existing systems (e.g. a swarm of Reapers) through to building in larger quantities a new generation of low-cost systems, designed specifically with the needs of the swarm in mind. Such a design must encompass the payloads and sub-systems as well as the airframe to fully realise the cost savings.
The science of swarms
The ‘science of swarms’ is founded in the lessons of nature. In classical terms we can characterise them as having four major characteristics(2):
1. The swarm consists of independent and autonomous individuals
2. Individuals in the swarm must act reactively
3. The swarm is self-organising (and simple rules can lead to complex emergent behaviours)
4. The swarm optimises itself for the good of the swarm, not the individual.
The UK MoD 2011 Joint Doctrine Note(3) concluded that ‘Unmanned Aircraft Systems (UAS) have already changed, and will continue to change, the way that we conduct warfare’, identified ‘Swarming/networks new ways of working’ as an opportunity and, in one illustration predicted fully autonomous swarms by the 2020s. The associated high level of autonomy does not however sit as comfortably. In addition to identifying a number of ‘weaknesses’, including ‘public perceptions’, that ‘the Law of Armed Conflict (LOAC) may constrain high levels of automation/autonomy’ and that ‘UAS are potentially vulnerable to cyber and communications link attack’ the document notes that: ‘Associated technologies are developing at an unprecedented rate and the relentless nature and speed of these advancements make it hard to assimilate, analyse and fully understand the implications … our conceptual thinking lags behind the art of the currently possible, never mind what lies in the future …’
Thus to overcome some of these hurdles we could, freed from classical biological constraints, pursue a less prescriptive view and consider ‘swarm-inspired’ or ‘hybrid-swarms’. Now we have started to define what our swarm might look like and how it might behave, the question is to what extent they can address the needs and challenges of the 2020 battlespace…
One notable need, as highlighted within the UK 2010 SDSR(4), is that the future operating environment will ‘place a premium on particular military capabilities, including intelligence, surveillance, target acquisition and reconnaissance (ISTAR).’ This desire for information is shared by other nations(5,6,7) and spans the full extent of the physical and electro-magnetic spectrum. Such desires however place a significant burden upon the available ISTAR assets and those who interpret their outputs.
The UAS swarm offers the intriguing concept of ‘distributed sensing’ — across spatial, temporal and spectral domains. Indeed it is useful to think of our swarm as being defined not in terms of physical proximity but on the principles of co-ordination, shared purpose, mission and intent. Any needs for spatial proximity will ultimately be driven by the capability of intra-platform datalinks or the link to the ground control station(s). The constant ‘chatter’ arising from this cohesive communication may prove to be a limiting factor (cost, power, range and advertising of presence). Thus the swarm might do well to mirror the natural world, where individuals use the principle of stigmergy (‘indirect communication by modification of a common environment’(8)) to co-ordinate indirectly. For instance, ants do not ‘talk’ to each other but use pheromones which affect the behaviours of others. Significant research has been undertaken into the use of ‘digital pheromones’ or ‘shared blackboards’, which can maintain a history of activity(9) or a dynamic vector map on which to base decisions(10). This digital environment data also allows for dynamic role assignments, swarm level optimisations(11), or can also be used to provide feed-forward information. Swarm members can update the digital environment which other members can then react to.
Intra-UAS communications can focus on distributing data with an appropriate timeliness, e.g. it may be sufficient to know that 50% of the swarm is planning to head back to base in the next three hours (a fact which is likely to change quite slowly), without needing to know the complete route plans of the UAVs (which will change rapidly). Through implementing this architecture in a ‘broadcast’ like manner and effectively aggregating information flows we are able to reduce the overall communications overhead. Moreover, such a system is adaptable to limits in network capacity — low bit rates gives you basic performance, higher capacity allows increases in granularity for improved performance.
Two sensors are better than one
With the communications burden potentially eased, let us consider the concept of a UAS swarm, its members equipped with low-cost sensors and spread out in time and space, such that they form a ‘sensor net’ covering over a region of several tens of square kilometres. Through dynamically ‘tuning’ the position of the individuals and the operation of each sensor and then collating and integrating each ‘take’ of feed, we envisage the formation of a virtual sensor capable of a number of feats.
Electro-optical (EO) sensors inspecting a target from different angles could allow a 3D image to be generated. Equally, electronic surveillance (ES) sensors could achieve multiple ‘cuts’ from different direction thus enabling accurate position fixes, while synthetic aperture radar (SAR) images from multiple viewpoints could be combined to provide a better overall image and reduce shadowing effects. Furthermore, if different members of the swarm carried different sensors then, though diluted, this virtual sensor could collect data across both the EM spectrum and a large region of interest.
Through staggering the take-off times of part of a (sufficiently large) swarm and forming a ‘conveyor belt’, sensors could watch over a region for a suitably long period of time. Platforms returning to base could signal the upcoming replacements crucial cohesion information about the swarm, or ask them to relay the need for a different payload to be installed on the next batch to be ‘fed into the belt’.
The dynamics for such a concept could be defined using the swarm principle that simple behaviours can give rise to surprisingly complex goals (the flocking goal of birds and fish can be achieved with only three simple behaviours(12)). For instance, through a single and simple rule such as ‘ensure all UAVs closer to the base than you have less fuel’, freshly launched platforms would automatically slingshot deep into the field, others who see their reserves growing relatively smaller would naturally drift back towards base.
Behaviours can be defined so as to be complimentary or conflicting — in which case the dynamics of the environment dictate the resolution of the conflict. Platforms in a swarm may have one behavioural drive to maintain a degree of separation between neighbours, and another to loiter over a target, of interest — the more relevant the target the stronger the urge to loiter. These two behaviours should serve to maintain coverage over a wide area but enable natural clustering around the most interesting targets.
Other behaviours can be introduced to give UAVs tendencies to perform certain actions, move particular ways, seek out particular targets, group together in certain configurations, have activation thresholds, be more or less altruistic, sensitive, cautious and so on. The swarm may be made up of individuals with heterogeneous behaviours and capabilities, so that they self-organise into nodes, perhaps with sensors on the edge and communications links in the middle. Behaviours themselves could be passed around, replicated or shared, such that a ‘commanding node’ is always active, but the individual it resides upon may often change.
There is almost infinite flexibility in choosing a set of behaviours, however, the difficulty is that there is no known analytical method to calculate the optimal behaviours set for a particular mission goal. Research is ongoing, but at this stage heuristics and intuition are the best guides. Swarms are complex systems and so to some extent it is the management of this chaos into some useful purpose that is the art of swarm engineering (13).
We noted earlier the need to consider mechanisms other than full autonomy and indeed it is possible for our swarm to have the capabilities discussed without requiring its members to be wholly autonomous. In this hybrid approach, while individual platforms operate autonomously for large portions of a mission, the operator chooses to intervene where necessary (perhaps in more complex or sensitive situations) and take direct control. In this instance, the platforms not under direct control could be issued the behaviour to ‘fall into line’ and react appropriately to the piloted platform. For instance, the operator could take charge of the lead platform and initiate a change of heading to avoid a potential airspace issue, with the rest of the swarm happily self-organising and following suit as the mother steers the children out of harm’s way.
A less direct, and lower burden, control mechanism that applies to all members of the swarm would be to get ‘hands on’ with the digital pheromones. Imagine an interface whereby said pheromones are visualised through a dynamic and interactive contour on a digital map. ‘Hotspots’ could represent flight routes, the density of sensor coverage, or the importance of regions not yet explored. Through simple mouse clicks or gesture control the operator could adjust the density of these hotspots or add ‘super pheromones’ (irresistible to swarm UAVs), thus strongly influencing the individual and collective behaviours. Want the swarm to return to base? Simply fill the airstrip with super pheromones…
The needs of the many…
The UK’s Future Character of Conflict paper(14) highlights a number of issues, not least that the future battlespace will be contested. There is also a growing school of thought(15) that the air superiority enjoyed in recent operations certainly cannot be guaranteed against a more symmetric threat. Early UAS may have been envisaged as being able to be put ‘at risk’, or indeed as being ‘dispensable’ as necessary. While UAS operational losses are not insignificant, it is claimed(2) that on a per thousand flying hour basis they are similar to those for manned platforms. The reality is that the complexity, cost and limited numbers of these (in demand) systems has meant any losses are significant.
Through its greater numbers and principle of individuals automatically reacting to their environment (dynamics and change is designed in at the basic level), our swarm offers a potential solution through an inherent robustness. The distributed nature of the swarm’s capability means, for example, that a malfunctioning sensor will ‘gracefully degrade’ the quality or coverage, rather than lead to a ‘cliff-edge’ loss as would be the case with a single (or even pair of) traditional platforms. Indeed, in these situations, greater autonomy offers a major benefit in that the swarm can automatically self-adjust without the need for an operator to micromanage/redistribute waypoints and tasks.
Cyber operations will undoubtedly be a key feature of the world of 2020 and the interconnected and semi-autonomous UAS swarm has the potential to be a prime target — however, initial research(16) is already being conducted to examine how such vulnerabilities might be identified and ‘designed out’ before systems are even built.
This natural robustness makes the swarm a good candidate for operations in contested airspace — perhaps even in the presence of capable air defence units (ADUs). When engaged by such a system, individual members (perhaps those least necessary to the mission) could offer themselves up as sacrificial systems, while the rest head for safety. Equally, a swarm equipped with suitable weapons and/or electronic capabilities could take a more offensive stance through sheer numbers and/or in a manner similar to existing systems such as MALD(17).
While the swarm offers clear potential military benefits, we must again consider whether the levels of autonomy discussed are too far removed from current capabilities, even for the 2020 timeframe. Shifting the boundary between human and autonomous control is a combination of technological, cultural and doctrinal challenge. How much decision-making are we prepared to let autonomous systems do? A suitable vision of the future may be one where, starting at the most basic level, we allow more and more decision-making to be devolved to the platform but always retaining the ability to take over when needed.
One approach may be to incrementally introduce the characteristics of the swarm to current ways of working. Today’s UAVs often switch between being directly controlled and, when appropriate, autonomously navigating to operator-defined waypoints. With the right supporting technologies, allocating a second or perhaps a third UAV to the operator is certainly feasible,(18) although great care must be taken over the cognitive loading.
The use of greater autonomy in instances where platforms are in a ‘lower risk’ (or ‘dull’) phase of the mission is perhaps where the swarm can start to be explored. For instance, in the conveyor belt scenario, platforms in the arguably less complex ‘transit’ part of the conveyor belt could autonomously fly within defined corridors and navigation rules, while the operator directs their UAVs at the sensing end or those about to land. The rest of the platforms could share information among themselves in order to make sure that no more than two of their number needed attention at any one instant.
We leave you with the analogy of a young teenager learning to drive for the first time and getting in a car with dual controls and a highly experienced teacher. In the early days, the teacher is doing most of the work and ensuring safety while the student is concentrating like crazy on simply turning the steering wheel. As the student develops, they start to take on the ability to also control the clutch and so on. The teacher still occasionally has to step in but is now primarily relegated to a supervisory role. Eventually the student is capable enough to go out on their own and tackle unexpected situations. Control and responsibility has been gradually shifted from the teacher to student so that at all times the system remained safe and capable. Our UAV swarm is eager to learn but when will we be prepared to teach it…?*Daniel Page and Dr Howard Tripp are senior consultants at Roke Manor Research Ltd (a Chemring Group Company) carrying out research and development of advanced technologies for unmanned systems.
1. Melling, Michael & Kumar, GRASP Lab, University of Pennsylvania, https://www.grasp.upenn.edu/
2. E. Bonabeau, M. Dorigo, and G. Theraulaz, Swarm Intelligence: From Natural to Artificial Systems. 1999: Oxford University Press.
3. Joint Doctrine Note 2/11 – The UK Approach to Unmanned Aircraft Systems, Development Concepts Doctrine Centre, UK MoD.
4. Securing Britain in an Age of Uncertainty: The Strategic Defence and Security Review, October 2010.
5. The US Navy’s Vision for Information Dominance, May 2010.
6. An Air Force Strategic Vision for 2020–2030, John A. Shaud, General USAF Retired, Adam B. Lowther, Spring 2011.
7. Joint Operations for the 21st Century, Australian Defence Force, May 2007.
8. David Keil and D. Goldin, Indirect Interaction in Environments for Multiagent Systems, in Environments for Multiagent Systems II, 2006, Springer.
9. H.V.D. Parunak, M. Purcell, and R.O. Connell. Digital pheromones for autonomous coordination of swarming UAVs. Proceedings of 1st AIAA Technical Conference and Workshop on Unmanned Aerospace Vehicles, Systems, and Operations, 2002, Norfolk, VA, USA.
10. C. Zhang and R. Ordonez. Decentralised adaptive coordination and control of uninhabited autonomous vehicles via surrogate optimisation. Proceedings of American control conference, 2003, Denver, USA.
11. H.V.D. Parunak, S.A. Brueckner, and J.J. Odell. Swarming coordination of multiple UAVs for collaborative sensing. Proceedings of 2nd AIAA Unmanned unlimited systems technology and operations aerospace land and sea conference, 2003.
12. C.W. Reynolds, Flocks, herds and schools: A distributed behavioral model. Computer Graphics, 21 (4), 1987.
13. C. Castelfranchi, The theory of social functions: Challenges for computational social science and multi-agent learning. Cognitive Systems Research, 2 (1), 2001.
14. Strategic Trends Programme: Future Character of Conflict, Development, Concepts and Doctrine Centre, February 2010.
16. Analysing Autonomous System Architectures for Security Properties, Conmy et al, SEAS/TR/2012/1, April 2012.
Aerospace International Contents - July 2012
News Roundup – p4
Farnborough preview p 11
2012 Farmborough Air Show
Foward looking p 16
Selex-Galileo radars and sensors
The perfect swarm - p 18
Intelligent autonomous UAVs
Winning the X(WB) factor - p 22
Report on Airbus Innovation Days
Setting the standard - p 26
RAeS input for new international standards for helicopter simulation
Letters - p 29
Developing technology to intercept rogue aircraft
Plane speaking – p 30
Interview with David Hess, President, Pratt & Whitney
The last word – p 34
Keith Hayward on UK national aerospace planning
This is a full article published in Aerospace International: July 2012. As a member, you recieve two new Royal Aeronautical Society publications each month – find out more about membership.