Why Don’t We Have Swarm-Robotics Yet?

We're far away from seeing a large swarm of drones

‘Swarm-robotics’ is a branch of robotics that deals with lots of individual identical robots that work together to complete a task. For example, in a search-and-rescue mission, imagine sending out a thousand tiny quadcopters that all ‘talk’ to each other. Those drones would all be communicating with their nearest neighbors to figure out what to do. The idea of having a swarm of robots is useful for plenty of problems; and in-theory is a very good solution. But for some reason we don’t really see very many swarm-like robotic systems in the wild.

To be clear, we’re not talking about multiple different kinds of machines working together to perform a task. We’re talking about identical machines working together to perform a task as a group.

The biggest reason this model never caught-on is because robots are still expensive. With most robotic systems, you only have to worry about the cost of a single robot, or usually a handful of robots. But purchasing a thousand identical robots can be very costly. Swarm robotics becomes more useful in numbers, so the price of individual units will need to come down drastically for it to make sense financially.

Another reason swarm robotics hasn’t caught-on is because machine-to-machine communication technologies aren’t that great at scale. Technologies like WiFi and Bluetooth work fairly well, but every once in awhile the technology fails. This isn’t a problem if you’re connecting your smartphone to your car stereo. But if you have a thousand machines all communicating with each other, there is a lot of surface for failure. A one-in-a-thousand failure rate will become a near 100% failure rate. Since this is such a vital piece of the swarm-robotics model, machine-to-machine communications will need to become more reliable.

The algorithms used to control swarms of robots are very complicated and take a lot of time to develop. The algorithms that individual robots use to achieve their goals are already very advanced. But robots that are members of a swarm require another set of algorithms designed specifically to exchange information with other members of the swarm, and to figure out how to collectively complete an objective. Right now the algorithms required for this exist, but are very challenging to implement. This is the single most important (and most difficult) hurdle to cross.

At least three things will need to change for swarm robotics to make sense. Reducing individual robot costs, improving machine-to-machine communications, and commoditizing swarm algorithms are all very important to the development of swarm robotics. Until then, we won’t see swarms of drones flying around in the sky.

1 Comment

  1. Interesting. However, I think that swarm technology is closer to realization than you make it sound. The examples you use are for completely new types of technologies. This keeps the entry bar quite high. That being said, if swarm robotics is required to take existing technology to the next level it will result in more (corporate) interest. A specific example would be autonomous vehicles. The eventual goal is to have all vehicles communicating with one another to allow for optimal performance(drag reduction) and hazard alerts among other benefits. Tesla has already implemented the first step towards this goal with its cars shared learning. As they drive new roads they upload their data so other cars can access it when needed. I don’t know how much of this is in real time, but one can clearly see the direction Tesla is taking. In short, I believe we are within 5-10 years of the first true swarm robotics (software) technology especially with a radical innovator such as Tesla in the market.

Leave a Reply