(Originally written for PREDICT/Medium in 2018)
People have an affection for dogs for many reasons: it could be an emotional support pet, a lapdog or a watchdog.
The latter is the focus of this article, yet no doubt even emotional support is fairly within the scope of deployed AIs.
So, the watchdog is an animal which can perform four functions if it is ‘deployed’ to guard your premises.
- It could prevent intrusions by its mere presence. The more frightening it looks and the more actively reacts to movements around the perimeter the better.
- It could detect intruders if they decide to give it a try. A dog does it by using its superior (well, superior to the humans, that is) sensory capabilities, like hearing and smelling.
- It could ‘report’ its findings to the controller. In other words a dog would bark and consequently alert the owner to the impending situation that is an intrusion.
- It could physically interrupt the break-in process by attacking the intruder. The repertoir of a dog to thwart an intrusion ranges from physical presence to the snap-and-zap typical of predatory animals.
For these capabilities they require basic anemities like food, water and a little cuddling.
So, in order to build a machine that could perform the same tasks it has to have these abilities plus more in order to be feasible as a replacement.
So, let’s start building!
By definition a land-based, self-propelled autonomous machine equipped with sensors and means to subdue an intruder is an Unmanned Ground Vehicle.
Our version will be a Totally Truly Autonomous Machine — UGV. And we will call it DogBot! henceforth.
There MIGHT be fully-autonomous or semi-autononous UGVs in various stages of development, but it is reasonably assumed that such machines aren’t deployed yet since the even UAV drones aren’t autonomous which reached their usefulness zenith already.
Contemporary UGVs are mostly of military nature and consequently are up-armored.
Also, the UGVs are generally propelled by petrol or diesel engines.
These two attributes determine their weight and consequently their size. Their size usually is smaller than a little car, but their weight is more than that. This allows for military UGVs to carry some kind of armament as well, usually an electrically fired machine gun and an electrically fired automatic grenade launcher. They aren’t particularly fast or agile nor able to traverse distances more than a couple of kilometres.
A second class of UGVs are born out of the need to defuse or destroy IEDs from a distance. These are usually remote-controlled self-propelled robots with a mechanical manipulator arm and occassionally with some kind of an electrically fired firearm to shoot trigger mechanisms of IEDs to bots in order to render them useless. This firearm usually is some kind of a shotgun.
The third class is the experimental cargo and reconnaissance UGVs of many shapes and forms. The current trend shows that this category will be very wide very soon — just take a look at how self-driving cars are reaching developmemt milestones. And self-driving cars are, by definition UGVs of the experimental cargo class.
So, if we are to find a viable replacement of our trusted and domesticated fellow, we need to do something else, namely to create a purpose-built TAM.
The dog is an agile animal, capable of hitting the speed of 45 mph( 72 km/h) (greyhounds racing) and having an endurance of keeping the quick pace for a few miles miles.
Its obstacle-crossing abilities are dependent on the actual shape, strength and flexibility of a given dog, but it is fairly possible for a fog to cross trenches, climb slopes (the steepness of the slopes is however a major factor), jump over barriers (height being a restrictive factor), push away objects (weight could be a factor) and to ford trough water bodies.
To achieve at least the same level of agility we need to find a suitable platform. For simplicity we don’t list all reasons what we need to evaluate, but a PMBOK-approved (as in Project Management Body of Knowledge) end result would be an electronically directly controllable quad bike equipped with an electric engine.
It has roughly the some weight, size and agility and a good all-terrain traversing ability.
It is able to reach speeds more than a dog could, and it could sustain high speeds as long as there is enough energy (fuel). It could cross obstacles a dog could, it could push away objects like a dog could and some models could even ford the water at the same speed a dog would do it (like the Gibbs Quadski).
But it can’t jump. This is going to be a shortcoming, but only until you supplement the homeguard system with an integrated UAV
But — a DogBot! can have a hardcoded map database, so it knows how to get to particular spots within the premises.
A dog is a quite unprotected animal, since it lacks armor (unlike for example turkeys having some kind of an armor). A dog compensates for this shortcoming by being proactive and resorting to attack instead of a passive defensive posture.
An DogBot! could be more armored than a dog, but let’s see what are the needs. It needs to be protected in a way that if kicked it should be painful and useless and it has to be able to withstand repeated blows by blunt objects (like a tree branch or a monkey wrench, etc.) Protection from firearms — in this context — couldn’t be an issue here, since that would be in the domain of combat TAMs, which are based on completely different priciples.
A dog might be visible for the passers-by or there might be a sign to warn potential tres(s)passers that a dog is located somewhere in the premises and might be roaming free.
As with dogs a dogbot could be visible or there could be a notice on the fence warning against setting a foot on the private property. It might or might not be enough to scare someone off.
A machine with even outdated sensory equipment could detect humans better than a dog would. It could use motion sensors, IR sensors, object recognition or the combination of any suitable sensory input.
Motion sensors, night vision-capable motion capture, night vision-compatible vision / cam + recorder, GPS, embedded geoinfo (map, dbase), and even a capacity to see trough walls.
Reporting / warning capabilities
A DogBot! could be designed to have an output device in the form of a speaker, so it might even bark, but it is more important that it could have an almost infininite number of output communications devices. It could send an email or SMS to the owner, stream realtime video to the local police, or even feed the intruder a live voice response from the law enforcement.
Getting physical: the legality
Now, this is a delicate issue, so let’s try to look at it from a legal angle first.
I THINK that the main question here is that what is allowed legally for the property owner if confronted by an intruder.
If it is legal to employ an unsupervised and freely moving dog without a muzzle, then — even though there will be much debate because of this — if the UGV causes less or equal amount of damage in the intruder, its use will be permitted eventually.
If in some juridictions it is legal to use lethal force against someone entering private property (without the need to establish their intentions first), expect the DogBots! pop-up there FIRST.
Even though the question of indirect responsibility will remain a factor (like responsibility for richoceted bullets, or using lethal force against minors), some armed UGVs will probably persist in around expensive households, key installations and at places of residence where the owner could do so.
And let there be no doubt: the one who could disassemble and assemble the trigger mechanism of a semi-automatic firearm or a car engine’s block could make it. Without much hassle.
Getting physical: in practice
I think that the encounter logic shouldn’t be tuned to seek a head-to-head situation. The best encounter logic would be to let the DogBot! stay at about 5 meters away from the intruder, from which distance it could use its loudspeakers, light omitting sources, could stream a recognizable video, and if the need be could deploy non-lethal defences, like CS-gas.
Our experiments show that there are numerous ways to implement DogBots! in the security field, so watch this space for updates!
(If you would like to contact for more specific information on TAM UGV DogBot! development and deployment — contact andras.korvin(at)longdawn.com)