Robots learning teamwork. HUAR!!

[Will]

http://www.engadget.com/2006/06/14/r...sing-imminent/

[DanClass]

Originally Posted by Will

http://www.engadget.com/2006/06/14/r...sing-imminent/

thank you and

fuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuck YOU!

[teerts]

Pasted from the article:

If robots ever hope to rise up and enslave their human masters, it's going to take no small amount of teamwork to get the job done, and luckily for our future overlords, DARPA's shelling out serious loot to endow them with just the tools they'll need. The agency's latest foray into robotic empowerment comes courtesy of researchers at the University of Pennsylvania, who recently demonstrated a platform that allows multiple heterogeneous bots to communicate with one another and use a sort of AI "group think" to find and presumably terminate specified targets. In a beta test at Fort Benning's mock urban landscape, the Penn researchers deployed four so-called Clodbuster autonomous ground vehicles along with a fixed-wing UAV overhead, and tasked the team with using their cameras, GPS receivers, and wireless radios to identify and locate a series of bright orange boxes. Unfortunately, after the successful completion of their mission, the bots decided to hit up the base bar to celebrate, where after several drinks they reportedly went AWOL and were last spotted attacking orange traffic cones in downtown Columbus.

The real article: http://www.newscientisttech.com/article/dn9328

A team of autonomous flying and ground-based robots have successfully cooperated to search for and locate targets in the streets of an urban warfare training ground in the US. The system could help in search and rescue efforts and military operations and even has the potential to include humans in the team.

Researchers at the University of Pennsylvania, US, tested their system of team-working bots at a realistic urban warfare training ground at the US Army's Fort Benning base.

They hid bright orange boxes in the streets between buildings. An autonomous robot aircraft with a wingspan of 2.5 metres, and four autonomous ground vehicles in the form of modified model monster trucks, called Clodbusters, then set out to pinpoint the boxes locations.

Both types of bot carried GPS sensors and looked for the targets using colour video cameras. The Clodbusters used stereo cameras to judge distance, while the plane used a single camera. The robo-team members stay in touch via radio or Wi-Fi.

"Using ground-based and aerial vehicles means the team can take advantage of each of their best capabilities," Ben Grocholsky, a roboticist who worked on the project, told New Scientist.
Sharing information

The plane can rapidly scan an area for the target, but because of its height can only localise it to within 6 m of its actual position. It can also have problems with buildings obscuring its view. The Clodbusters provide local precision, and a different perspective. They can use information from the plane to find the target more quickly and then mark its position to within 0.2 m.

You can see a video showing the robotic team locating targets during the trial (60MB mov format). The trial was part of a larger project on robot teams, funded by the US Defence Advanced Research Projects Agency (DARPA).

No computer or a human need take a leadership role in the hunt meaning the bot team can still track down targets effectively if it should lose any one robot. Each robot is loaded with a map of the area, and is on a personal mission to find any targets inside it. When it has any information on where one might be, this information is also passed to other robots in the team so no data is held centrally.
Scaling up

"If the air vehicle sees something, then the ground vehicles are aware of it and will decide whether to investigate," explains Grocholsky. "Each robot has an idea of what it thinks is going on, in terms of probabilities of a target being in a particular place."

The researchers hope this distributed model will scale up easily, so that large networks of many different robots, sensors and even humans could be patched into a team.

"A key advantage of this approach is the way the information available to the robots is anonymised," says Grocholsky. There is no need for complicated coordination of the different elements of the team each just uses any information it gets to help with its own goals.
Map makers

"If a human was provided with a way to provide information in the right way, they could help out," says Grocholsky. For example, a handheld device that let a human make their guess of where the target was could send information in the correct form to the robots.

"There are many things I like about this," says Paul Newman, who works on robot navigation at Oxford University in the UK. "For one, it's pretty hard to get robots to cooperate when they're all the same. Finding a way to do it with ones that are different is impressive, as is having it work outside in the real world."

Newman's own research involves trying to have robots navigate entirely unaided. Adding that capability to a system like this would make it more valuable, he adds. "These robots start out with some idea of what is around them, robots that can build their own maps as they explore would make a team like this more flexible."