UAV swarm, anticollision - faa - detect-and-avoid

Since UAV activity is largely constrained by the see-and-avoid issues;

I though it would be appropriate to propose a solution which is compatible for every form of flight from gnat-weight to jumbojets. I've worked this out is a simulator - but here is the jist:

This requires only a single frequency and very short blips.

1. Every plane determines it's location and vector by GPS.
2. Then it transmits this information using one short blip every second or so.
a. The blip timing is based on current GPS location, so that the blips cannot interfere, and their location in Timespace tells us where they are coming from. Since GPS provides a very accurate, shared time base. So let t in microseconds = latitude\1 + longitude\1 * 100 + Alt in Km *1000 .
b. Odd seconds communicate vector (ie destination), Even seconds communicate Location.

3. There is a third variable required - which I call "Density". Each plane calculates the perceived Density of their airspace and sends that as well. Your density is your passenger count divided by the distance between you and other planes multiplied by their perceived density.

4. For manned aircraft nothing further is required - this beacon signal will ward off all unmanned vehicles.

5. For Autopilot or UAV, it is possible to chart a new safe vector by trying random vector changes and calculating their future density based on the vectors and densities of nearby airplanes. The vector which optimizes progress toward destination with lowest density is the vector chosen.

simple calculation requirements (ie Arduino), cheap transmitter/receiver requirements, and low power requirements. cheap unit cost, and no ground control.

Such a system could be demonstrated by coordinating a swarm of Ardu-Planes.


Views: 2265

Reply to This

Replies to This Discussion

The ACO math will be more complex in 3D, but I don't see that as a fundamental limitation. I was thinking about how you would program in stationary obstacles as well - this reminds me of Craig Reynolds' "Boids" -
I should note that the simulations communicated actual coordinates; and I have not simulated the blip-blip method of propagating signals; Your point is well taken, if 20uS is noise; then 100uS is 5:1 s/n; At 100uS, one can communicate a grid of 10000 cells per second.

If one communicates a large grid in odd seconds, and a more detailed local grid (say 5 miles) in even seconds, then 10000 grid points should be quite accurate. Since manned airplanes should never be within 1000? feet of each other, 500 feet is a suitable resolution. A grid of 500 ft*70 cells per side = 35,000 feet per local grid edge (plus altitude).

One could use altitude as a sub-resolution signal
for close planes, it would be more accurate; while for distant planes it would be more course:

series: GPS t = #.0 sec
t+0: Cell 1/1: alt 1000ft, 2000ft, ...50000 ft.
t+2000uS Cell 1/2: 1000ft, 2000ft, ...50000ft.

t+9998000uS Cell 50/50: 1000ft, 2000ft, ...50000ft.

Of course, one could simply use xbee, establish serial com, and send real gps coords. That I think would require more energy - but still not a great deal.

If it is not a certified system it will not be accepted by the other airspace stakeholders, FAA or any other CAA. Other systems exist but none yet are certified, certification will/would take years.
Every GPS form factor - including a geoTrex or a Garmin hand-held, or a gps-enabled cellphone is capable of being made to transmit this ping.

The problem is already demonstrated to be beyond the reach of our existing systems.
How much fuel and time are wasted because the FAA (system) is crap?
Least fuel routing? FAA = NO. ATC of parachute avoidance? NO. ATC of glider avoidance? NO. Localized peer-to-peer avoidance is the only solution which can integrate light-flights such as micro-UAV and parachutes with every other type of flight.

I doubt that Human Pilots are more capable than this system.
As an RC pilot, and occassional small plane passenger, I have some experience in the "real world"

People, as it turns out are very poor a concentrating, they tend to perform worse, not better under pressure, and they can experience "brain farts", or similar, "I have no idea what I was doing?" No system which relies on Humans can certify the humans to be awake, sober, friendly, competant, or non-heart-attack prone, or not-distracted. Nor can it provide for location awareness on foggy days; They cannot manage ground traffic well. Humans do not track multiple bodies well.

Under what scenario will 13 human pilots in a 5 mile grid outperform this algorithm?

I hear you, but I think you can change the definition of "certified" This algorithm can be certified in ways that humans cannot be certified. Any system which is widely adopted - would be embraced by stakeholders if it is technically and cost effective.
Just implement the transponder into strobe lights used for high towers and building (over 300 feet or whatever) that solves the problem.

You can also avoid fixed obstacles with way-point planning.

This brings up another issue - If the FAA is permitting "homing devices" to be exported because they do not permit unchecked waypoint programming - we may find a market for pre-fixed devices. For example, a farmer may order a cropcam which is pre-programmed for his field - since he cannot change the waypoints - the risk of mis-use is largely mitigated. He might even qualify as owning his own airspace up to 500 feet for example.

I came to the same conclusion about stationary beacons, but you typed faster ...

Look at this as a generalized tagging device - you could even employ it to define "no fly" zones. The real issue becomes how you deal with objects that aren't tagged, which leads to passive detection schemes as the backup. Even before trying to take on the FAA with this approach, there are other markets where the operator fundamentally has control of his local airspace (e.g. cropcam, military uav's, etc).
when a predator has a servo falure a human has to land it . i dont know of any AP system or any other computer that can deal with THAT level of unknow intereaction . if someone has created a computer that comes even any ware close to human intelligance please let me know . it sure would be fasinating . sure computers are better at boring ,dirty, high reaction time type computing but can a computer realize its not very safe to fly in that close proximity to other craft and avoid the situation all together? only if you told it to . computers cant yet think for them selves . and we are not just talking about one computer avoiding another computer but a computer avoiding ALL potential collisions . that cannot be done with any known system at this are absolutly right in saying an AP can do better then a human in many respects but a human will for a long time to come be better at dealing with the unkown and unexpected. my proof of this is the fact that there is not one truely autonomus system being produced by anyone right now . every uav on the planet right now must have humans overseeing them.
Too much to answer, it wasn’t the FAA but commerce that allowed the return to launch sale overseas. The new ATC will be more efficient, and lastly a system (collision avoidance), does exist in Europe that they use in manned gliders. It’s reactive in the sense that each glider must have one on board for the system to work. It also has the position of ground based obstacles as well as gondola wires programmed into it. It’s used in the Alp’s and I’m not sure if it is certified.
Many RC wings already have 2 ailerons per wing, or 4 control surfaces; in most cases (dead servo near neutral) the other control surfaces could pick up the slack.
A rudder very often can substitute for Ailerons and vs.versa. I read that one University had worked out a self-aware uav that could handle these issues. Arduino could do it with an infrared leveler.

This algorithm can do extra-regional coordination, meaning that it's "Density" messages will be propagated from plane to plane such that the liklihood of congestion is minimised across a larger region than a single-hop would define. Humans can't do that without ATC, and don't do it well with ATC.

To your final point, I'm proposing we address that void by implementing a "swarmable" uav brain - the Arduino is capable - it needs a gps with an interupt on the GPS second hand. And we need a cheap direct-drive radio/receiver (rc radios are basically direct-drive).

We don't need human intelligence, we need sensors and communicators and an algorithm for avoiding collisions. Humans are bad at collision avoidance (bats on the other hand...)..

actually chris has misgivings on the ability for arduino to handle the extra load however these are exactly the kinds of thing im going to be working on with my basic-x platform ,(32bit,60mips) with future plans to move to a netburner CPU later . i am already working on autonomous soaring and have even achieved a 20 min unpowered flight . the idea is that you would have each unit in the swarm looking for sources of lift and as sources are found of course the data would be passed to the other swarmbots . don't get me wrong i fully agree with what you are saying but the technology will not be consumer ready for a while .
A 60-72MHz ARM7 might be adequate. An LPC2103 costs around $3 in moderate volume. GPS chipsets in volume are probably in the $15-$25 range. The radio is a bigger question - if you go with a chipset rather than a finished module, it could be sub-$10. However, for prototype and initial production, XBee is probably a good starting point.

It's an interesting concept, and something that might fit in the DIYdrones arsenal or alternatively something I might be motivated to pursue separately. Send an email if you want to discuss offline.

Reply to Discussion


© 2019   Created by Chris Anderson.   Powered by

Badges  |  Report an Issue  |  Terms of Service