Slashdot Mirror
Pheromone Robotics
An Anonymous Coward writes: "This is the official text I just came across: "The HRL Pheromone Robotics program aims to provide a robust, scalable approach for coordinating actions of large numbers of small scale robots to achieve large scale results in surveillance, reconnaissance, hazard detection, path finding, payload conveyance, and small-scale actuation." But it's the spooky image that grabbed *me*..." Here's some more on the pheromone-sniffing robots pictured, and some more information about making robots that hunt in packs. The page has not been updated for a while, but it's worth seeing.
Ants are very simple creatures that working collectively in groups to accomplish fairly complicated tasks using mostly smell.
Modeling behavior along these lines and allowing simple creatures to relay very simple state messages with each robot repeating it to others would allow behavior and information to be propagated and acted on even in hostile situations. (Only short range communication is possible for instance.)
ie If the robots are searching for something and one finds the target it could alert the others around it and they could repeat the message and alter their behavior accordingly - if required. Eventually the alert would filter across all robots and reach "home" at which point a response could be propagated back to the successful creature the same way.
If your creatures are too simple there are limitations, however. If you put certain acids on ants other ants will assume they're dead - the smell trigger - and carry them to "dead ant" pile even if they're struggling. The "un-dead" ant will be carried back to the dead ant pile repeatedly until the smell wears off.
=tkk
Bill Gates - Creationist?!?
Imagine hacking these robots just by farting near a swarm of them ;)
Israel A. Wagner's home page about Ants, Robots and Computation is here and it's a great and interesting compilation of data on this topic. Absolutely recommended.
For more information on the ability to accomplish complicated tasks with simple rules take a look at this book The Computational Beauty of Nature.
Very informative book, lots of good explanations, diagrams, and the code for his software is available on the website. As a plus he seems to have written the book using free software which he acknowledges at the end of the book). His programs run under linux. He has some very well done graphics (even some dual-image stereograms) which were created with gunplot. I highly recommend this book.
Now if only Tom Ridge could figure out which end is up...
And Bush would come clean about the "pretzel" incident.
"It is a greater offense to steal men's labor, than their clothes"
Here's a section from the article that explains basically what I just tried to explain.
I don't mean to be flame bating or anything, I just thought the writeup was a bit confusing/misleading...that or I'm just an idiot and didn't understand what they submitter meant.
Basically its a ton of cameras.
Use visual recognition to detect stuff like troop movements....
Or just use cameras+ human watching + set coordinates already.
Picture vietnam movies, with the guy radioing for a mortar strike.
Now picture no one in harm's way, but the cameras and coordinates for mortar strikes are at Mr. God's hands.
Later: Camera + visual recognition = Missle Guidence system
Later: Camera + High powered AA guns + some elementary physics = anti aircraft guns
But here's the kicker... If you network ALL the cameras, so you know information everywhere, then you can calculate things better...
Its like automagic driven cars, the more networked, the more you see around JOE BOB in his big ol truck... Because of networking alerts you to whats around it... The more you know.
God spoke to me
The idea for the pack hunter sounds like a basic neural network with BEAM robotics. You can get a fairly good idea what BEAM(Biological, Electronic, Aesthetic, and Mechanical) is all about at http://www.solarbotics.net .
Basically, the philosophy of BEAM is that all robotics can be made from imitating the form of nature, and all can be made from the same basic components. You start by replicating single cell organisms and insects, and eventually progress to neural networks. The cool thing about BEAM robots is that they're entirely automated. You don't have to program their behavior, they works by "instinct."
If you want to make your own BEAM robots, just read some of the tutorials on http://www.solarbotics.net . Just don't be tempted to start with a complex robot. The idea behind BEAM robotics is to start simple, and work to complex. All you will need to start is a basic understanding of electronics and some cheap electronic components, which can usually be found at RadioShack(or by tearing apart old household appliances.)
I have no desire to reach nirvana.
The "pheromone" business is no big deal. That's how creatures too dumb to make maps mark their world. The territory is the map.
Can you imagine a cluster of these babies!?
I never thought that would actually come up seriously.
Dacels Jewelers can't be trusted.
Quote "But it's the spooky image that grabbed *me*..."
Yeah, a picture of little robots is real spooky. I bet this guy's shadow gives him a heart attack.
What the hell are you talking about sir? Did they say they're mounting mines and missiles on these little guys? I guess I missed something here.
I am aware of this project quite a long time, since I'm working on a similar project called "Swarm-Bots" [web site www.swarm-bots.org]. According to me their use of the term "pheromone" is not more than a catchy adjective to label their work. The research, as displayed on their web site, does not take many ideas from the ethological studies of ant colonies. For instance the robots communicate directly with each other, NOT through the environment, which is what ants use pheromone for. To me, it is merely an integration of the dynamic programming technique with mobile robots coupled with VR display interface. For those new to the subject, there is a new approach, called "swarm intelligence" that aims to create intelligent systems from a group of distributed simple agents. An excellent description of this approach is available in the "Swarm Intelligence: From Natural to Artificial Systems By Eric Bonabeau, Marco Dorigo, and Guy Theraulaz" . In this approach, the agents communicate through the environment, called stigmergy, to achieve group level tasks. There is no centralized control, yet the whole system is very scalable and robust. I hope to report some news on the progress of the Swarm-Bots soon.
Gladly. Let me sum up ten pages of buzzword bingo bullshit in a few simple sentences:
"We've got some robots and some ideas, neither one of which we've actually implemented yet. We only have one robot right now, but, here's a nice time-lapse photo of the same robot scooting around a corner to make you think we own dozens of them. Due to the incredible design of our software that doesn't exist yet, they communicate with you and eachother using radios that dont exist yet and draw stuff on the walls with an IRLEDs that don't exist yet either. We're burying the idea in a sea of buzzwords because we need to secure funding..That, and we're afraid youll find out these ideas have been around for years."
Next.
Bowie J. Poag
So you know, the HRL webpage has been updated.
She sat at the window watching the evening invade the avenue.