Swarming robot boats demonstrate self-learning

The sea-going robots are made using digital manufacturing techniques

The sea-going robots are made using digital manufacturing techniques

(Credit: Biomachines Lab)

From Giz Magazine

   FEBRUARY 2, 2016

Robots may be the wave of the future, but it will be a pretty chaotic future if they don't learn to work together. This cooperative approach is known as swarm robotics and in a first in the field, a team of engineers has demonstrated a swarm of intelligent aquatic surface robots that can operate together in a real-world environment. Using "Darwinian" learning, the robots are designed to teach themselves how to cooperate in carrying out a task.

A major problem facing the navies of the world is that as ships become more sophisticated they also become much more expensive. They are packed with highly trained personnel that cannot be put at risk, except in the most extreme circumstances, and even the most advanced ship suffers from not being able to be in two places at once.

One solution to this dilemma is to augment the ships with swarms of robot boats that can act as auxiliary fleets at much lower cost and without risk of life. The tricky bit is figuring out how to get this swarm to carry out missions without turning into a robotic version of the Keystone Cops. The approach being pursued by a team from the Institute of Telecommunications at University Institute of Lisbon and the University of Lisbon in Portugal is to rely on self-learning robots.

Led by Dr. Anders Christensen, the team recently demonstrated how up to ten robots can operate together to complete various tasks. The small robots are made of CNC-machined polystyrene foam and 3D-printed components at a materials cost of about €300 (US$330). The electronics pack include GPS, compass, Wi-Fi, and a Raspberry Pi 2 computer. However, the key is their decentralized programming.

"Swarm robotics is a paradigm shift: we rely on many small, simple, and inexpensive robots, instead of a single or a few large, complex, and expensive robots," says Christensen. "Controlling a large-scale swarm of robots cannot be done centrally. Each robot must decide for itself how to carry out the mission, and coordinate with its neighbors."

Instead of using a central computer or programming each robot individually, the swarm operates on what the team calls a Darwinian approach. In other words, each robot is equipped with a neural network that mimics the operations of a living brain. The robots are given a simple set of instructions about how to operate in relationship to one another as well as mission goals.

The robots are then allowed to interact with one another in a simulated environment and those that display successful mission behavior are allowed to proceed. The "fittest" robots from the simulations are then tested in the real world.

According to the team, the clever bit about the swarm is that, like schools of fish or flocks of birds, none of the robots know of or "care" about the other robots beyond their immediate neighbors. Instead, they react to what their immediate neighbors do as they determine the best way to fulfill their mission objectives such as area monitoring, navigation to waypoint, aggregation, and dispersion. In a sense, they learn to cooperate with one another.

The team is currently working on the next generation of aquatic robots with more advanced sensors and the ability to handle longer missions. Eventually, they could be used in swarms numbering hundreds or thousands of robots for environmental monitoring, search and rescue, and maritime surveillance.

The team's research is being peer reviewed and is available here.

The video below describes how the sea swarm works.

Full article here

Views: 1391

Comment by Hector Garcia de Marina on February 4, 2016 at 1:10pm
I was reading the paper and or they are very naive or very pretendeous by stating that they are the very first ones using swarm of autonomous outdoors robots? Literally on the text u.U

Admin
Comment by Thomas J Coyle III on February 4, 2016 at 1:18pm

@Hector,

Yes, definitely pretentious.

Swarming autonomous outdoor robots have been around for a while.

Regards,

TCIII AVD

Comment by Pierre Kancir on February 5, 2016 at 12:36am

@Hector, Thomas,

Actually,they aren't much real swarming autonomous robot. Many demonstrations are using centralised computing and/or augmented localisation (vicon camera...) or don't perform swarm interaction (juste mouvement according predefined partern...) 

So it is a good step !

Comment by Hector Garcia de Marina on February 5, 2016 at 1:01am
Hi Pierre,

First of all you are right indeed. Nevertheless there are many outdoor robots in a totally distributed fashion. I am pretty confident about that since I am working in this field since ten years ago xD.

Do not get me wrong, their platform is very nice! It is just they way they use to present it .
Comment by Tobias Witting on February 5, 2016 at 3:00am

"CNC-machined polystyrene foam and 3D-printed components" and ... tupperware boxes. I like it.

Comment by Pierre Kancir on February 5, 2016 at 3:57am

@Hector,

I am doing a PhD on the subject, if you have reference/publication/video on real swarm, I am interested in !

Comment by Patrick Poirier on February 5, 2016 at 4:14am
Java based controlers makes it an different than the usual ROS-Python approach and the implementation of ad-hoc mode on the Wifi is one of the few demonstration of particular network configuration.
Comment by Hector Garcia de Marina on February 5, 2016 at 5:25am

@Pierre

sure, no problem. Maybe many of the things that I am going to post you know them already. If you need something else more in particular, feel free to ask.

First of all, nowadays in 2016, I would say that there is a big consensus about what is a controlled environment, from conversations with relevant people at important conferences such as ICRA:

- Software numerical simulation. 

- Indoor robot + "Optitrack".

- Outdoot robot + "GPS + Wifi com" .

Here by robot we mean something very well understood with very verified models, such as quadcopters or unicycles, e.g. wheeled robot. Researching groups in the field of control theory, IA, etc go for the second option mostly. It is fancier than the first one and much less resource (time, personnel, maintenance, money, etc) consuming than the third one. And nowadays nobody in the field will call "novel or exciting" to have a quadcopter or a car moving around if the main electronics employed are GPS + Wifi/Off-the-shelf comms in a sunny day. Fifteen years ago yes, today for sure no. So controlled indoors are the most convenient option, and not because the outdoors version is more challenging, it is just more expensive and many of the top researching centers are focused on developing theory (the real added value and not for instance just to design out of the blue some cost functions).

However, for new "exciting" technology for outdoor robots I will refer you to the prestigious "Journal of field Robotics" (JFR), where they state clearly in their webpage, "JFR will not publish articles in which the experimental validation is restricted to simulation or controlled laboratory experiments".

Regarding distributed control theory, the field is very active nowadays, specially in the topic of Formation Control. Mostly mathematics but they give detailed analysis about topics such as convergence, switching sensing topologies, harsh communication constraints (btw broadcasting continuously a wifi/radio signal is not what you want in a distributed environment). Just a quick search in google scholar about Formation Control and you will find a very active topic.

Regarding actual researching groups, the Australian National University has a strong group for teams of UAVs or ENAC in France as well. For marine vessels, you have the group of Prof. Fossen in Norway and CalTech, and for teams of underwater vehicles the facilities in Bremen are simply awesome. And here I am missing people from St. Petersburg, a very strong university in both, theory + practice, but it seems that they are a bit hermetic.

Btw, regarding the original post. When I was studying my master's degrees in 2010, me and a friend developed a very similar system (at that time with ARM7 controllers!) of small vessels with Radio+GPS+Compass. I remember that we were also playing with many distributed approaches. Here you have a recent example  https://www.youtube.com/watch?v=xZCa2ylfmaI

Comment by Erik J on February 5, 2016 at 7:14am

The research paper and code is worth a read, I like the engineering minutia they bring up (could be more I guess) https://github.com/BioMachinesLab/drones but that hasn't been touch in a while though... So I wonder if they are going to run something next year as well.

@Hector Thanks for all the references, might be really interesting stuff, but it is expensive knowledge you are linking :-)  

http://www.journalfieldrobotics.org/  is a paid publication ~40 dollars for downloading an article.

Prof. Thor Fossens university courses seems to have a nice technical level as well. Literature is +$100 though

http://www.itk.ntnu.no/emner/ttk4190

But great post!

Comment by Pierre Kancir on February 5, 2016 at 8:19am

@Hector !
Thanks for the references, I didn't know the Journal of field Robotics, but I will definitely look at it.

What make me very sad about swarm/multi robot robotics is that it stay mostly in research field.... We get plenty of control algorithms etc but no real platform deployed or used.... Research is good, but if we can used it for real application, it is better / or on real world robot!

Well, I hope the project I work on will succeed in demonstrating another real swarm with real system (and be use after that!) 

Comment

You need to be a member of DIY Drones to add comments!

Join DIY Drones

© 2019   Created by Chris Anderson.   Powered by

Badges  |  Report an Issue  |  Terms of Service