Autonomous anti poaching drone

Hi All, it was suggested that I post about a project I am funding where we have designed a long range patrol drone that for use in areas of South Africa where Rhino poaching is affecting private reserves. We have built a quadplane Skywalker x8, with off the shelf components, and we will supply these to local reserves at our cost price.

The scope of the project is as follows.

We have a highly specced patrol UAV that will be able to autonomously identify poachers on the ground via a FLIR camera and a University Deep Learning Project. For phase one we still need to do two things and hopefully members here may know how we can do it which will mean funds can be invested in the project and not programming.

The UAV has a pixhawk 2.1, rasberry Pi3b and intel Movidius stick. What we are going to implement is a hot area into the video frame so that when a contact moves into this area the UAV will do one of 3 things:

  • Move the Tarot Gimbal to keep the object in frame

  • If multiple objects move in a manner to preclude a gimbal move, zoom out the camera or if not possible, increase the UAV’s altitude.

In addition to this once the contact is confirmed to be a poacher, the UAV needs to loiter around the position and change the centre of loiter as the objects move.

Ok that’s the easy part now the hard part.

It can take time for ground assets to respond so we have a 2nd quadplane UAV armed with a dye delivery system. What needs to happen is that we need to allow these to swarm but they need to get continuous GPS updates from the patrol UAV as to the location of its centre of loiter.

Once they arrive on scene they need to use their onboard cameras to move onto target and then loiter while at the same time avoiding each other. EDIT I do realise that this technology can be weaponized but we have committed to the university supplying the detection software that it never will be and in this case it could actually save lives.

Anyone have ideas on an easy way to do this?




Great to see drones being used for anti-poaching but the delivery of dye onto the poachers makes me worry. I think it would be much better to leave out this feature.

The problem is that once the poacher is identified, they can easily slip out of the reserve and blend in with the local towns folk. Once they are dyed they can’t and it makes apprehension easier if they do evade capture.

I share Randy’s concern, this is not a technological but an ethical problem.
Once a configuration like this one is builded and published, there is no garantee that it might not be used in a lethal scenario, from a third party.
There are legal issues as well, most counties have strict rules concernig shooting projectiles of any kind, from a flying vehicle. This can end up in court with criminal charges, here in Canada there is a very well knowned case: "Roman Candle Attack Drone 2.0"

Cool project. Poaching is out of control and puts alot of live in danger. I agree with the comments from Randy and ppoirier - not a good idea to deliver dye from a drone. But why is this a requirement? Why can’t you maintain 100% visual contact with the suspected poarcher as he tries to retreat from the scene? Just keep the camera pointed on the suspect and follow his movements with the drone. This also provides irrefutable evidence to the courts.

An easy way to deconflict the drones is with altitude. Have one drone flying at 300 ft and the other at 400ft, 100 ft of separation should be more than eneough to compensate for any barametric or GPS altitude differences between the two drones. This is easy to implement, does’t require drone communication and is nearly fool proof.


1 Like

Haha, love it. It would be super interesting if you’re able to give more details of how you’ve done this, would be fantastic. It would be a great blog post here.

With regards to concerns about the dye pack, I can understand the the reticence of others to personally help make that happen but I don’t see why it’s something that should be actively excluded from conversation. Using dye to identify poachers is a lot more peaceful than what usually happens to them on the ground with people involved if caught, and may actually save lives on both sides.

The ‘dual-use dilemma’ is not a new one, but it could be argued that there is already explicit support for technologies that could be used for far more nefarious purposes:
^^ chemical weapons delivery agent
^^ bomb delivery agent
Not to mention DO_SET_ROI for surveillance, etc etc

The nature of this area of technology means almost all of it falls into dual-use, if we restrict what is discussed/developed on that criteria then it would severely limit what is allowed. A lot of the projects I hope to release this year will be very useful for research and conservation, but could equally be very useful for hunting and poaching, something I have been very aware of when designing them.


putting these dilemma aside, seems prettty clear to me that your developpement on MAVERICK is taylormade for this type of mission. If that can can save a race from extinction in a ""civilized’’ manner, please take the lead.

I do realise the ethical implications of this which is why our UAV’s will never be armed but can see that if we publish this it could be misused. Having said that our VTOL Skywalker x8 could already be militarised but we will never go there.

The issue is that the vegetation is very dense and once the poacher leaves the reserve, we cannot follow. If they are dyed the police can move into the townships and arrest them.

We haven’t done it yet, I know what we need to do, just trying to figure out how much is already in arduplane that can be leveraged and how much we need to code.

Have you first really done the numbers on flight time? Not what a certain quad plane can do but a real scenario given the distances of the park, the average acquisition time, typical winds and the hover time needed after acquisition. Where are your dye drones located. How long on average for them to reach the patrol drone. Can you really differentiate between other animals and poachers? How fast and accurately could they deploy a dye pack to a moving target. How much time do they have before they are shot out of the sky. My gut is telling there is no way this can be done without a hybrid fuel aircraft and even then unlikely given the current state of the art. High altitude surveillance from a non-quad plane hybrid powered maybe.

We have a number of landing areas that the patrol UAV can go to for battery changes during each patrol and we anticipate that it can fly for around 2.5 hours between each change. The UAV will be configured to take off after each landing climb to a specified altitude and transition to horizontal flight at which stage it will loiter until it reestablishes telemetry and then will resume its patrol. We don’t ever hover, other than take off and landing but rather will loiter when needed.

The idea is that there will be two patrol drones so one will be airborne at any one time. The response drones can reach the patrol drone in under 10 minutes and are located around the reserve.

We are still working on the dye bombing so I can’t say at this stage how accurately they can lock on but given the terrain, a target can’t move that fast.

The identification system we are using has already achieved a high degree of success but we also have human verification via a central control station.

So this is a very small park.Like 5 square miles.

It’s a number of privately owned reserves that collectively make up a larger single reserve unit. In this case the collective reserve has only one vulnerable border, over which incursions into the individual reserves occur, so the response drones will be located along it. This system isn’t designed for reserves the size of Kruger national but rather the smaller reserves that surround it and share common borders with it.

Well the single border limits the problem. But I would be very surprised if a Skywalker can get 2.5 real world hours with quad motors and a gimbal hanging off the belly. I assume you will need to Patrol 24/7 so you will need staff to land in the dark and swap batteries every few hours. It would be worth doing some tests first as to the flight time of the Skywalker X8 with the Tarot attached to make sure it meets the requirements. Is a foam plane going to survive being opened up 12 times a day?

Not dising your idea just the practical engineer in me looking at the problem.

Basically need to patrol at night only in a 6 hour window. Skywalker won’t last forever but we don’t intend it to, airframe gets replaced at every service cycle. We are about to commence flight testing.

Hi Robert, I spent 2 years flying Pixhawk equipped anti-poaching fixed-wing drones in the KNP (Olifants/Talamati/Pretoriuskop/Phabeni), Manyeleti, Hluhluwe/Imfolozi, etc in 2014-6 with thermal cameras so have a bit of experience in this field. Not to put you off but some of our biggest problems were the following:

  1. Poacher detection: I flew 290 logged hours and 17400+km and I saw just 1 poacher group (we flew 330ft altitude, 43km/h, viewing ~18Ha a minute, 2hr endurance per battery). No apprehensions or arrests for our entire team after 1000+ hrs.

It is extremely difficult to find 2-3 persons in 1000’s hectares of African bush, on open plains or canoes on water yes, otherwise the calculated chance of detection is about 1-2%. Even flying fencelines with alarm triggers we still failed to see anyone besides anti-poaching personal.

  1. Difficult conditions: Area too large, too vegetated (thick bush), too much time (with only 1 drone), 6 hrs of night flying very taxing on operators, poachers come when you sleep/rest or before or after you fly. Air temp at midnight in Feb 2016 was 37°C (100°F) thus poachers become invisible to the thermal cameras. Also hectic storms, wind, insects, dust, rain, humidity, etc

Once you’ve overcome those then you have:
Poacher retention: It is almost impossible to get a apprehension team to the poacher undetected, (hence your question about dye marking).

The results in the end did not justify the means. Not until the drones can do the following without human intervention: charge/refuel, launch, search, retain, shoot, etc.

Hi Graham, one of the criteria is that we reduce the manpower required to operate this program. To do this we have teamed up with a university with an established AI project that has already been tested in the Kruger. We expect the hit rate to initially drop where we will fly until the system learns to deal with the terrain.

Heat is an issue with the FLIR but we also have IR onboard with 940nm LEDs that can illuminate objects up to 120m away.

The man power to replace the batteries is also reduced as we have tried to make it as simple as possible but all your concerns have been factored into what we are trying to address.

If this were my project I would look to using Foxtechfpv’s new 6 hr flight time hybrid quad. No changing batteries at night patrol, can hover and follow targets directly without having to fly a circular loiter pattern. Tough plastic body rather than foam. More rain, bug and dust resistant.

Solves a lot of mission critical problems.

Problem is it will be too noisy, we have already had to take steps to minimise noise as far as possible to overcome resistance to UAV usage. You would think that this would be easy but the amount of resistance and red tape due to abuse of cheap drones for game watching is astounding. There are two similar projects to mine, both commercial and independently we all settled on an X8 VTOL.