Study Finds That Self-Driving Cars Have A Lot To Learn From Ants

You know who has their shit together? Ants. You don’t hear about ants doing all kinds of hand- (or antennae-) wringing about how they should best do something, or getting into fights about morality or culture or whatever. They just do ant shit. Sure, their culture isn’t super deep and their music sucks and individually they’re maybe not terribly attractive, but there still may be a lot we can learn from the ants. Especially when it comes to self-driving cars.

Yes, that’s right! It’s not all just dorks into coding and crypto that can add something to the development of automated driving, but also ants – at least, according to a study published this January in Transportation Research Interdisciplinary Perspectives  titled “ANTi-JAM solutions for smart roads: Ant-inspired traffic flow rules under CAVs environment,” by Marco Guerrieri and Nicola Pugno. The study found that ants were incredibly good at avoiding traffic jams, even in high-density/high-traffic contexts.

Want to read the abstract of the paper? Sure you do:

The behaviour of ants has inspired various scientific disciplines due to their ability to solve even complex problems. During their movement, ants generate trail networks that share many characteristics with vehicular traffic on highways. This research aims to estimate the values of traffic flow variables (mean speed, density, and flow) in ant trails without intersections or branches that could alter the dynamics of each ant. A case study in an outdoor environment was analyzed. The macroscopic traffic flow variables of interest were estimated using the deep learning method and the YOLO detection algorithm. The results show that ants adopt specific traffic strategies (platoon formation, quasi-constant speed and no overtaking maneuvers) that help avoid jam phenomena, even at high density. Emerging technologies, including smart roads, communication systems, and Cooperative and Automated Vehicles (CAVs), allow us to speculate on the use of traffic control systems inspired by ant behaviour to avoid the risk of congestion even at high traffic volumes, as demonstrated by the preliminary results of this research.

Now, one thing worth noting about this overall approach to self-driving that differs from the self-driving technologies you may be familiar with (Waymo, Cruise, and Tesla’s Full Self-Driving/FSD, even though FSD has yet to be deployed in an unsupervised, non-Level 2 context) is that this paper focuses on self-driving systems that communicate with one another and incorporate “smart roads” and other tech where the infrastructure that these automated vehicles use is an inherent part of the overall system that allows these vehicles to drive independently.

Oh, and if you’re curious, the “YOLO detection algorithm” refers to an object detection algorithm known as “You Only Look Once.” That one was new to me, too.

Most current systems just rely on the car’s own sensors and cameras and on-board computers to navigate and drive, where a system like the one being described here is much more holistic, with other cars and infrastructure all communicating to provide a much more robust understanding of the world at large that can help inform the automated vehicle how to drive.

So, how do ants fit into all this?

Well, I’m not sure if you’ve noticed, but ants are sort of autonomous vehicles themselves, and they tend to move in some very large groups, while still maintaining orderly progress. If you don’t believe me, head to a local park or sidewalk and drop a delicious jelly donut on the ground and just wait. I suspect it won’t take long until thousands of ants have established a few highways to that jelly donut and back to their jelly donut processing facilities in the ground. [Ed Note, or Ed Tip if you prefer: Bring two jelly donuts, so you can eat one. Why torture yourself? – Pete]

The study found a lot of fascinating details about the rules by which ants handle traffic, the key ones of which they outline in the study:

Ants can solve complex traffic control problems by simple rules;

• the relationship between the flow rate q and density k revealed no evidence of jam phenomena;

• in contrast to the typical shape of fundamental diagrams of highways and other road infrastructures under uninterrupted traffic conditions, the mean speed of ants appeared to be quasi-constant with density. In particular, at low ant density, the flow q increases linearly with the density k, whereas at high density, it reaches the maximum value (i.e. the capacity) and remains constant;

• spatial distribution of individuals along the trail shows that ants move predominantly in platoons;

• within each platoon, ants maintain almost equal speed, keeping small time headways;

• no overtaking maneuvers are performed.

This is pretty remarkable, and with these rules in mind, when I think back to ant highways I’ve watched, they seem to fit what I remember: ants tend to move in big groups (platoons, as they call them here) and at a remarkably constant speed. Also, individual ants tend not to pass one another.

Of course, if you’re interested, you really should read the whole study, which is, as I’m sure you suspect, vastly more complex and in-depth than the barest of coverage I’m giving it here.

Ants “drive” in ways fundamentally very different than humans tend to, because I think ants are collective-minded as opposed to individual-minded, like most of us human drivers are. When it comes to automated driving at scale, it’s important to realize that perhaps we should not be attempting to just replicate human driving behavior; maybe part of the advantage of taking a human out of the equation is that we no longer need to be bound by human rules and preferences when driving.

When we drive ourselves, of course we want to pass people sometimes. It’s fun! Is it the best way to avoid traffic jams? Hell no. But if the human is not in the equation, what sense does it make to keep making the mistakes we make? Maybe all individual-focused automated driving is a dead end, really, and collective, collaborative, and highly communicative automated driving makes more sense?

I mean, if we’re going to willingly give up the experience of driving, at least for some of our automotive travel, why not let the machines do it the most efficient way possible? And if that means driving like ants, then, well, so be it.

Of course, doing something like this would require buy-in and standardizations agreed upon by manufacturers, infrastructure development, governments, and maybe more.

Also, we should probably get the input of some ants, too.

Yet Another Study Shows That Driver Assist Systems May Be Doing More Harm Than Good

Volkswagen Seems To Be The First Automaker With A System That Gets A Car Off The Road If The Driver Doesn’t Respond

‘Fully Automated AVs May Never Be Able To Operate Safely’ Says One Of The Oldest Professional Computing Technology Organizations