9. • Air Traffic Control
• Complexity Challenge
• Always Improving
CAFE 2015, Santa Rosa, California
CENTRAL MANAGEMENT
RESOURCE LIMIT
10. • Sidestep Complexity
• Pilots
• Car Traffic
CAFE 2015, Santa Rosa, California
DISTRIBUTED MANAGEMENT
UNLIMITED NUMBERS, BUT…
11. • Distributed
• Simple, Universal Rules
• Successful Autonomy
CAFE 2015, Santa Rosa, California
CARS - UPSIDE
FAIRLY LARGE NUMBERS
12. • Traffic Jams
• Defensive Driving
• Unpredictable Stopping
CAFE 2015, Santa Rosa, California
CARS - DOWNSIDE
UNDESIREABLE CONSEQUENCES
13. CAFE 2015, Santa Rosa, California
DISTRIBUTED MANAGEMENT IMPLICATIONS
SAFETY LIMIT
14. • Distributed Coordination
• Fluid Motion
• Swarm Research (Flocking)
CAFE 2015, Santa Rosa, California
PART II – SWARM SCIENCE
SOLUTION FOR CROWDED SKIES
15. CAFE 2015, Santa Rosa, California
NATURE’S EXAMPLE
SOLUTION FOR CROWDED SKIES
32. • Lift Sharing
• Dynamic Soaring
• Personal Robot Swarms
CAFE 2015, Santa Rosa, California
COOPERATION
MORE EFFICIENT FLYING
33. “The More The Merrier”
CAFE 2015, Santa Rosa, California
SWARM SCIENCE
Tyler MacCready, PhD
Ocean Lab
505 E. Wilson Ave.
Glendale, CA 91206
tyler@oceanlab.com
Editor's Notes
Swarm Science is the Study of Collective Behavior. It’s a multidisciplinary field covering biology, chemistry and physics. And it’s also relevant to engineered systems. A few years ago I started a company, Ocean Lab, that applies Swarm Principles to the management of large numbers of aquatic vehicles. Today I’ll be talking about applying these principles to aviation.
There are three parts to this talk. First I’ll describe how if the CAFÉ vision succeeds it will lead to a big increase in the number of planes in the air which will then have us bumping up against limits of resources and safety. I’ll describe how Swarm Science has solutions to these limitations. And I’ll finish with some examples of how group flying can actually be beneficial.
The CAFÉ Foundation has a vision for the next generation of efficient personal aircraft. We don’t know exactly where this design space will end up, but one key target is Quiet Short Take Off and Landing.
With quiet planes, urban airports become more viable. Brien has been pushing this notion of Pocket airports in recent years, as a way to make shorter-range air travel so convenient that it becomes a reasonable option for commuting.
So if the CAFÉ vision succeeds, if we get our new generation of personal aircraft and Pocket Airports to fly from, then for many it will be a preferred commuting solution and there will be widespread adoption. This means a big increase in numbers of planes flying.
How big an increase? No one knows, but let’s speculate for a moment. Right now there are about 20 GA airports in the Bay Area. If we increase this tenfold we get more like the number in all of California. And if they really become preferred commuter ports it’s easy to imagine ten times the traffic at each airport. This means that in the future skies of the Bay Area we could see ten times as much traffic as all the small aircraft flying in all of California today. It is not hard to imagine seeing thousands of planes flying some of the preferred routes at rush hour.
It could be a lot of planes. Right now we’re thinking about designing individual planes. But if it works then what we are really designing is a crowd, in which case there will be management challenges.
So say the airplane technology is worked out, the pocket airports are in place, what are we facing with this management challenge? I want to look at this in terms of limits, and at the heart of this discussion is the distinction of central management versus distributed management. We’ll need both. Central management is when information from the group is gathered together in one place where decisions are made and commands are sent back out . Distributed management is leaving the decision-making in the hands of each individual who then only concerns themselves with their close neighbors.
For aviation, Central Management is Air Traffic Control. When planes are close enough to interact with each other the complexity of this control task grows exponentially with each additional plane. Air Traffic Control already takes a lot of resources in terms of people and computer power. I’m not going to talk more about central management today except to say it will always be important, and we will always be operating at the limit of our resources.
Fortunately, with airplanes, we have the option for Distributed Management as well. We can sidestep much of the complexity issue by having an active pilot or sophisticated autopilot in each plane. Distributed Management is fully scalable. It’s a great way to accommodate very large numbers. If we want to see where this might be headed for airplanes we can take a look at car traffic.
Cars show the amazing functionality of a distributed system. Each driver roughly follows simple, universal rules of the road, and a large number of commuters are able to get where they want when they want.
But car traffic also shows an important downside. At some level of crowding our reliance on Defensive Driving rules leads to unpredictable stopping, traffic jams. This is not desirable.
I could show a picture of a typical traffic jam, but I thought this was more interesting. This is an experiment conducted in Japan where each driver in this circle was told to keep going 20mph and don’t run into the person ahead of you. It works fine until a certain level of crowding is reached, at which point a traffic jam develops. For cars, stopping is a hassle. For planes, it’s entirely unacceptable. At some level of crowding, a distributed approach based on defensive rules reaches a safety limit.
This is where I introduce swarm science. Swarming is collective behavior that emerges out of distributed rules for coordination. If we have the wrong rules we can get things like traffic jams. What we really want for airplanes is a more continuous, fluid behavior. Research on how we might do this has been going on for quite a while, inspired by the natural flocking behavior of birds.
So let’s start with birds. Here are some local Northern California geese. Obviously they have no problems flying together. And even in very large numbers they don’t crash into each other or fall out of the sky. We face limits. Birds don’t face limits. In their flocking they are showing us the solution for crowded skies.
And I would be remiss if I didn’t show a clip of starlings. These ones in England are showing us some of the densest grouping around, and what’s amazing with these birds is that, unlike us and our car traffic, the more crowded they get the more fluidly the group behaves.
Nature is showing us a solution for safe, dense flying beyond the defensive distributed limit. How this is achieved is the study of Flocking Algorithms. Now I’ll show some early work in this field as well as some more recent results.
In 1986 Craig Reynolds pioneered the study of flocking by distilling it down to three simple rules. It’s a distributed algorithm where each agent makes its own decisions based only on close neighbors. The first rule, Separation, is basic defensive flying. It’s like feeling a buffer zone around each neighbor. But then he adds two more rules beyond defensive flying. One is Alignment, trying to match the average of the neighbors’ velocities. This makes the group fluid. And the other is Cohesion, an attraction toward the center of mass of the neighbors, which keeps the flock together.
Here are some clips from his early flocking simulations. He called the distributed agents Boids, and this shows the flock maneuvering around obstacles.
And here is a more recent simulation of the Boids algorithm. In this case the programmer has added a general group attraction to the green sphere, and a buffer zone to prevent collision with the red sphere. The important thing to notice is that the birds don’t collide with each other and they don’t stop.
Lots of people have studied flocking since that early work. One of the more prominent researchers is Tamas Viscek in Hungary. He and his students have had good success using the same three rules but making them distance dependent. The names are a little different but it’s the same general idea. An agent is repelled from close neighbors, attracted to more distant neighbors, and in between there is this orientation zone where they just try to align with their neighbors.
By varying the spatial relations of these rules they are able to create different group behaviors. In the example on the left there is no orientation zone, only attraction and repulsion. The result is sort of a mob scene, no collisions but also no alignment. In the middle example, with a small orientation zone, the agents begin to align into a circling flock. And on the right, when the orientation zone is increased, the whole flock begins to move in the same direction.
Recently these algorithms are starting to be tested in hardware. Here are Quadcopters flown by Viscek’s group. They are running their distributed flocking algorithm plus being given very simple, high-level commands, first to form a circle, then to line up, and finally to follow a target, this car.
Here is a Swiss group doing a similar experiment, this time with flying wings. Their flocking algorithm creates a circling group of non-colliding planes. (The black speck here appears to be a bird attracted by what looks to him like thermaling behavior.)
So this is where we are. These are real vultures, but they could easily be UAV’s running a Distributed Flocking Algorithm.
On to part three. I started on a negative note, with the notion that crowds are a problem, and then suggested that Birds and Swarm Science have a solution to enable a lot more crowding. But I want to finish more positively by noting that the flocking we see in birds is not just a safety behavior. Birds flock even when they don’t need to, when there are no predators, no feeding, no fraternizing. The bottom line is: Birds often just prefer to fly in groups. There must be added value to flying in a crowd.
One value might be situational awareness. If you are a creature flying through invisible air, one of your best sensors is to observe the behavior of other similar creatures. In this way your sensing radius is as big as your crowd. And for us humans with our radios we can even pass a lot more information than birds can. This added knowledge can result in safer flying.
We can also use that information from the crowd to improve our flying. Learning about local tailwinds, upcurrents, regions of lower turbulence, can all help us find better air. We may even get updates on special scenic routes to fly. In these ways group flying can be better flying.
And finally there’s the opportunity for cooperative flying, like the lift-sharing of geese in a V formation.
There are sailplane pilots these days doing the sort of dynamic soaring that Phil Barnes talked about this morning. They extract energy from turbulence and add it to their forward momentum by pulling various quick maneuvers. Vultures also do this. Getting some close-neighbor knowledge of upcoming turbulence could improve this sort of dynamic soaring.
And who knows, perhaps someday we might fly surrounded by our own personal robotic swarm.
Each of these cooperation possibilities can lead to more efficient flying.
So I’ll close with this message: Embracing Swarm Science not only can Enable the Future’s Crowded Skies, it may be that Flying in Flocks becomes Preferable, and in the future we’ll take-off singing the mantra of the swarm: “The More the Merrier”. Thank you.