Heidi and autonomous herding

Heidi and autonomous herding

Heidi is the 19th century girl from the Alps who herds her beloved goats. Nowadays we are moving towards autonomous herding. This article summarizes the present state of autonomous herding and explores requirements for such a system.

The article which triggered everything

While going through the new articles displayed by my feed reader the following title caught my attention: Why Use Agriculture Drones? Main Benefits and Best Practices ([Al19]). Agriculture drones? I opened it and skimmed the content. The following paragraph, although very brief and only one of several Drone Best Practices in Agriculture presented in the article turned my imagination loose:

In livestock farming, drones can keep an eye on the cattle as it grazes on pastures, reducing the need for human workforce on horseback and trucks. Using thermal sensor technology, drones can find lost cattle, detect injured or sick animals, and calculate their exact numbers. Admittedly, drones are capable of doing a better cattle surveillance job than herding dogs.

[Al19]
Theoretically, all steps of the dairy process could be automated.

I was beguiled by the vision created by this last sentence (my emphasis):

[…] drones are capable of doing a better cattle surveillance job than herding dogs.

Yes, drones could indeed herd cattle – in a fully autonomous fashion. Even more is possible, autonomous milking machines could do the milking with autonomous milk tanker transporting the milk to the dairies where automatic production processes would process the milk and its derivatives so that autonomous trucks can deliver it just-in-time to the relevant supermarkets.

Literature review of the autonomous herding topic

A brief review of some patents

Okay, so I had this vision of a radiant future or a nightmarish techno-hell whichever you favor, but I couldn’t have been the first to have it though, right? Right. So, I had a look at patents first, thinking that companies will most likely have an interest in implementing such a system.

There is this company Dinklage Feed Yards based in Sydney, which already patented an Unmanned Livestock Monitoring System internationally in the year 2017 (see [DFY17]) raising also a claim on

an animal locator and herding device onboard the unmanned aerial vehicle for determining the location and controlling the movement of livestock.

[DFY17] , p. 22

They’ve also made the next step in claim 13 by patenting the “controlling [of] a plurality of UAVs with animal locator and herding device(s)” ([DFY17] , p. 23).

The degree of automation in the dairy industry is high.

Same year, four months earlier the company Digi-Star has patented an Agricultural Drone for Use in Livestock Monitoring (see [DS17]). Noteworthy is a document from the year 2000 patenting An Unmanned Vehicle to be Used in a Stable or a Meadow (see [LRH00]). This vehicle has a “disinfecting means,” a “manure slide for removing manure which is lying on the floor” and even an “electric shock device” – whoa, spicy! Lely Research Holding, the company behind the unmanned vehicle, was granted two years later an US patent for a Construction for Automatically Milking Animals (see [LRH02]).

I could go on with citing patents, but you got the general idea. Needless to say, that this brief overview does not constitute any legal advice. If you want to build and sell something in this area you might want to check with your lawyer first.

A brief review of scientific literature

My second step was to skim through some scientific literature. I was a bit disappointed as I would have expected more material, but at least I could dig up some names and some studies. As I only devoted a very limited amount of time for this literature review, I can only hope to have found the most relevant information. Should I have missed any important studies, which is most likely the case, let me know in the comment section.

An „analogous“ dog helping out with a flock of geese.

So, the first project in this area seems to have been the Robot Sheepdog Project led by Richard Vaughan at the Oxford University between the years 1996-1998. Since this is more than 20 years ago, I would say this is a very early account of the idea, even though they created rather a duckdog than a sheepdog, controlling the behavior of a flock of ducks, rather than one of sheep. But the reason is quite understandable as

[…] it was soon decided that herding sheep was too big a jump in one step, and so the robot herds ducks instead.

[RSP-1]

In my opinion this doesn’t make the results less impressive. An interesting contribution from 2018 [PaChKiSh18] presents an algorithm for herding flocks of birds away from airports with the help of a drone without breaking their formations.

Various publications of Richard Vaughan in cooperation with other researchers exist, for example [VaHeSu97], [VaSuFrCa98] and [Va99]. A complete list of publications of the Autonomy Lab led by him can be found here.

A brief look at real herds

Although there seems to exist substantial interest in automatizing an increasing number of farming aspects like health monitoring of cattle, the interest in using robots to autonomously herd larger animals seems to be very limited.

How about regular people? How do herders, cowboys and their bosses feel about drone herding? I’ll leave this one to you, just search for drone herding on YouTube and you’ll see.

Drones already have various uses in agriculture.

Conclusions on the present state of autonomous herding

It’s time to answer the question asked implicitly in the title of this chapter. So how far are we?

With respect to autonomous herding I would say not very far yet, some interesting studies have been made, some prototypes exist, but that’s about it. And although there most probably is a lot of research ongoing right now, the problems posed by an autonomous system of such magnitude are – in my opinion – far too great to be successfully tackled in the next 3-5 decades to come.

But with respect to other aspects of an autonomous farm, technology is very advanced. Companies like Lely Research Holding but also various others didn’t settle for filing patents but implemented the systems they described, so we do have today more or less fully autonomous farms where robots feed, milk and monitor the cattle. And watching these on YouTube is pretty impressive. For the readers interested in more information: a more or less exhaustive account of companies in the field of DairyTech and their products and solutions can be found in this article.

Who cares about sheepdrones?

So, the next question is whether there is an economic interest in herding drones? I’ll ask back: Why not? If it will be per total cheaper than whatever will be in place at that time, surely. And it will be cheaper. Just consider the time necessary to train one single dog. And dogs are living social beings.

Imagine: you just increased your sheep herd by a few hundred sheep. In the days of sheepdogs, you would have to get also some dogs, and even if they are trained ones, give them some time to adapt to their new situation. In the days of sheepdrones, you just order two more of them and add them to your formation, they will start working instantly, no friction, no get-to-know-each-other time, no establishing a new hierarchy in the pack, you just need the electrical power to keep them up in the air, right? Hmm, but how do you get to charge those things? Do you suddenly need two more shepherds who constantly change the batteries of the sheepdrones?

This is where the other pieces of the puzzle come in.

The vision of autonomous herding

Will there be drones in the air and robots in the grass in 50 years?

Before I continue, let me write a short disclaimer: As a software engineer, I am well aware of many of the disruptions caused by technology on societal, cultural and individual levels. I am also well aware of many of its beneficial achievements. Whereto we go as a society is something, we must decide together through the mechanisms of our democracies. In the following I will completely ignore all ethical aspects posed by the problem at hand – although I am well aware of several of them – and describe the system solely from a technical point of view.

So, we are basically talking about a fully autonomous livestock monitoring and herding system, where human effort is limited to edge cases, like treating sick animals, patching technical problems and changes in the system configuration. We have already seen that automatizing lots of aspects of dairy farming have been successfully implemented. Autonomous herding seems to be dealing with various challenges though. In the following I will try to identify what is necessary for an autonomous herding system to work. Of course, such a system is only interesting for non-housed farms.

Requirements for an autonomous herding system

First let’s identify in more detail the tasks to be performed by such a system.

Basic requirements

  • The system is capable of herding a set of animals from the current location to another location. The second location is either autonomously identified dependent on various input parameters or set remotely by a human operator. Input parameters are the present location, weather conditions, state of the technical system itself, health condition of the individual animals, etc. Should this location be too far or should there be any changes in the input parameters, the system can adapt and define intermediary waypoints. The intermediary waypoints are based on previous configuration or on artificial learning.
  • The system should be capable of herding various types of animals like cattle, sheep, reindeer, horses, buffaloes, camels, swine, goats or ducks. The initial version will most probably work just fine with algorithms only for cattle and sheep.
  • The system must be capable to simultaneously herd the main herd and guide back potential strays.
  • The system is capable to identify the health state of each of the animals it is taking care of. For this each animal must have a unique identification tag.
  • Furthermore, the system is also capable to identify its own health state and to take countermeasures, should the health state be sub-optimal. The countermeasures can be either predefined or situation-based, according to a preset or learned configuration.

Extended requirements

Wolves and other predators can be a threat to the herd. Drones must deal with this threat.
  • The system is capable of working under various weather conditions. Should the conditions be of such nature, that the systems or the herds health is periclitated, the system should be capable to react with countermeasures to maintain the herds and its own health and safety.
  • The system must be able to deal with cases of animals who are so badly injured, that they cannot follow the heard any more. Even less badly injured animals whose state might worsen, must trigger a response of the system which is targeted at maintaining the health of the animal.
  • The system must be able to deal with predators threatening the health and lives of the livestock. These predators can be bears, wolves, tigers, lions and various other animals, dependent on the target region. Various responses of the system might be necessary in order to scare these predators off. The system is not allowed to identify people, other technology or even one of the herded animals as predators.

Security and safety

  • The system must be capable of protecting the herd against mischievous humans trying to harm or steal members of the herd.
  • Furthermore, the system must be able to protect itself in the sense, that it cannot be controlled by unauthorized personnel, any attempt on breaching, manipulating or hacking the system must trigger an alarm.
  • The system must be capable of dealing with partial failures and it should be capable of fixing minor issues with itself. The components should be able to assist one another in case of necessity.

Extensibility and interactivity

  • The system must be capable of learning, human intervention has to remain minimal should it be at all necessary.
  • The system must be remotely controllable, a remote human operator should have at all times at least a general idea about the herds and systems health. Limiting the amount of data sent is okay in circumstances where energy must be saved. Furthermore, the system must also be autonomously operational, even without connection to some remote servers.
  • The system is dynamically extensible and is also capable of detecting whether an increase in its components might be necessary for effectively performing its job.
  • The system should be capable of meaningfully interacting with similar systems herding a different set of livestock. If stray animals attach to a different herd, the responsible herding system should be capable of informing other systems of this fact. Furthermore, should multiple herding systems operate in the same area they should not get each other in their way.

Challenges of autonomous herding

Such a system poses several enormous challenges which we are still far from having tackled yet:

  • Developing effective herding algorithms for various species of animals, which can compete with sheepdogs.
  • Although drones can act in formations and prototypes of complex systems encompassing multiple actors exist, to my knowledge they still lack the necessary flexibility for a fully autonomous herding system by far.
  • Weather conditions pose a great challenge, dealing with heavy rain, heavy winds, unexpected snowfall or lightning are just the extreme cases.
  • Dealing with all kinds of terrains is also a big obstacle in the realization of such a project. While the system most probably won’t have to deal with movement through forests (at least not in its first version), movement in very steep areas, traversing of creeks and covering relatively long distances at high speed will be necessary.

Special challenge: Power consumption

And there is the aspect of power consumption. Power consumption will be high. Regular drones normally fly for far less than 30 minutes, also dependent on the number of sensors activated, the flying style and the weather conditions. Surely, batteries will most probably greatly improve with the passage of time. And of course, you can greatly improve the flight time through intelligent algorithms. I still believe that there will be necessary to also have large battery packs in the form of specialized robots following the herd who will recharge the drones when necessary. Even creating an infrastructure for providing energy and various services is imaginable. This infrastructure has to be, of course, entirely autonomous.

Conclusions – or what’s Heidi’s job in the future?

After doing a superficial review of filed patents and of scientific literature concerning autonomous herding, this article describes on a very high level the most important requirements for such a system. Although describing this system on a vision-level does not take more than a few sentences, materializing it will take at best several more decades.

Nonetheless, the progress made in the realm of dairy farming since Heidi’s days are astonishing. Where will she be, when the above-described vision will become reality? Most probably in front of some screens herding robots.

In a few decades we might have solely robots doing the herding.

But there is one more thing. Imagine you don’t need the whole robots-herding-livestock-thingy. What about just implanting minuscule chips which stimulate the animals’ brains so that it feels to them as if they would decide by themselves where to go and when to start heading there? Science fiction, you would say? Yes, it’s still science fiction, but we’ll get there, just have a look at this article.

Will Heidi then herd goats again?


References

[VaHeSu97] Vaughan, Richard; Henderson, Jane; Sumpeter, Neil. 1997. Introducing the Robot Sheepdog Project. In Proc. International Workshop on Robotics and Automated Machinery for BioProductions. Retrieved on 16.02.2020 from http://autonomy.cs.sfu.ca/doc/vaughan_biorobotics97.pdf.

[VaSuFrCa98] Vaughan, Richard; Sumpter, Neil. Frost, Andy; Cameron, Stephen. 1998. Robot Sheepdog Project achieves automatic animal control. In Proc. Int. Conf. on Simulation of Adaptive Behaviour (SAB), Zurich, Switzerland. Retrieved on 16.02.2020 from https://pdfs.semanticscholar.org/e6ab/17160acf25bfc2a22b1f27371edeb50ce2b4.pdf.

[Va99] Vaughan, Richard. 1999. Experiments in Automatic Flock Control. PhD Thesis, University of Oxford. http://www.cs.ox.ac.uk/people/stephen.cameron/sheepdog/vaughan_ras00.pdf

[RSP-1] Robot Sheepdog Project. Retrieved on 16.02.2020 from http://www.cs.ox.ac.uk/people/stephen.cameron/sheepdog/.

[LRH00] Lely Research Holding AG. 30.11.2000. An Unmanned Vehicle to be Used in a Stable or a Meadow. Patent No. WO 00/70941 A1. Retrieved on 16.02.2020 from https://patentimages.storage.googleapis.com/4a/5b/82/689ee2847d5236/WO2000070941A1.pdf.

[LRH02] Lely Research Holding AG. 15.10.2002. Construction for automatically milking animals. Patent No. US 6,463,876 B2. Retrieved on 16.02.2020 from https://patentimages.storage.googleapis.com/b3/89/4b/4aca1204738ebe/US6463876.pdf.

[DS17] Digi-Star, LLC. 30.03.2017. Agricultural drone for use in livestock monitoring. Patent WO 2017/053135 A1. Retrieved on 16.02.2020 from https://patentimages.storage.googleapis.com/e3/38/9d/7817182e06b5c5/WO2017053135A1.pdf.

[DFY17] Dinklage Feed Yards, Inc. 27.07.2017. Unmanned livestock monitoring system and methods of use. Patent WO 2017/127188 A1. Retrieved on 16.02.2020 from https://patentimages.storage.googleapis.com/74/8c/de/7bf96e79ba7463/WO2017127188A1.pdf.

[PaChKiSh18] Paranjape, Aditya; Chung, Soon-Jo; Kim, Kyunam; Shim, David Hyunchul. 2018. Robotic Herding of a Flock of Birds Using an Unmanned Aerial Vehicle. Published in: IEEE Transactions on Robotics (Volume: 34, Issue: 4, Aug. 2018). Retrieved on 16.02.2020 from https://ieeexplore.ieee.org/document/8424544.

[Al19] Aleksandrova, Maria. 2019. Why Use Agriculture Drones? Main Benefits and Best Practices. Retrieved on 16.02.2020 from https://dzone.com/articles/why-use-agriculture-drones-main-benefits-and-best.


You might also like the following articles:

Petre Sora

Petre Soras Interessen sind vielfältig und befinden sich an der Schnittstelle zwischen Mensch und Informationstechnologie. Als studierter Psychologe und Software Engineer war er sechs Jahre als Java-Entwickler in mehreren Unternehmen tätig. Mit der Gründung der Rezensionsplattform nososo hat er sich entschieden eigene Wege zu gehen.

Schreibe einen Kommentar