Picture walking up to a bucket of stuff in your garage and asking it to change the oil in your car. Then imagine that, in response to the command, a mass of self-reconfiguring robots moving around in the bucket climbs out, reconfigures on the fly into a shape that best suits the task, and proceeds.
That's the ultimate vision of Mark Yim, a researcher working on the latest generation of shape-shifting robots at the University of Pennsylvania. Instead of single-purpose robots, Yim and several research teams around the world are working on creating mutating machines made out of smart building blocks that can morph into different forms and perform a variety of tasks.
The key word here is versatility. In designing for unpredictable circumstances and strange environments, researchers say it makes sense to have a single, shape-shifting robot that can crawl through small holes, climb stairs, cross gaps or go through rubble. The robot's first mission is to save lives. Yim's robo-centipede will allow search-and-rescue personnel to look for survivors in a collapsed building by dropping the robot through a 4-inch hole drilled into the wall.
"When a building collapses from an earthquake or a bomb, etc., then digging for survivors is slow and dangerous for both the rescuers and victims," said Craig Eldershaw, a researcher at Xerox's Palo Alto Research Center, or PARC. "It's difficult to send in a standard robot, since no one knows what it's going to find there. The robot may have narrow pipes to crawl through, rubble to climb over, and some corridors clear of debris where it should try to make good time in moving. But a reconfigurable robot can change shape and adapt to the environment."
Hansel and Gretel, two bots that are being developed by researchers at PARC, use a combination of radio and ultrasound to help them form a map of an uncharted, ruined building as they look for potential survivors. Robin Murphy's team from the Center for Robot-Assisted Search and Rescue, which designed the tank-like Inuktun robots used in the 9/11 rescue efforts, is currently focused on enabling bots to determine their own configurations and choose the form that's safest for the situation. For example, a bot could automatically crouch when going over rubble.
Current demonstrations of the capabilities of shape-shifting robots have mostly focused on changes in locomotion, such as moving multi-legged robots into a snake, rolling them like a tank tread, or going from four legs to two. Daniela Rus' ground robots at the Dartmouth Robotics Lab self-reconfigure into a form that allows them to "flow" over rough ground like water. Other groups have demonstrated changes into useful shapes. For example, small reconfigurable robots could transform themselves into screwdrivers or hammers when needed.
However, most of these transformations have been done under human direction. For the bots to be truly independent, they need to be able to think, an issue that has been the subject of heated debate.
"It's a big question: Should there be a central brain or no brain at all?" said University of Pennsylvania researcher Yim. "There are people who are trying to develop systems which have entirely distributed intelligence -- all the intelligence is distributed over every module. There's another camp which says that you need to have some kind of brain or some kind of hierarchy. For example, if you look at your hand, you have fingers attached to a hand and a hand attached to an arm and multiple arms attached to a body. Similarly you can have one module for the hand which is the brain for all the fingers, and you have another brain at your shoulder which talks to the arms, and so on."
"Let's say you are in a room and there's a door and you are in the shape of a snake," said Yim. "But if you can't go under the door, you have to open the doorknob. The system might recognize all that and say, 'Oh, I need to be a human' in order to open the door. And it theoretically knows how to control the human shape, so then the problem becomes, 'How do I get from the snake to the human shape?'"
Current approaches use a set of known configurations, in effect a toolbox of shapes (a snake, spider or caterpillar, for example). The system figures out the best possible shape for the task and reconfigures the robot from one to another. A more difficult approach, which researchers are still working on, is to look at the task and have the computer automatically generate a shape to do the job.
Pictures:
Source: http://www.wired.com/news/technology/0,1282,65866,00.html?tw=wn_1techhead
"Terminators are coming! Terminators are coming!" :P
That's the ultimate vision of Mark Yim, a researcher working on the latest generation of shape-shifting robots at the University of Pennsylvania. Instead of single-purpose robots, Yim and several research teams around the world are working on creating mutating machines made out of smart building blocks that can morph into different forms and perform a variety of tasks.
The key word here is versatility. In designing for unpredictable circumstances and strange environments, researchers say it makes sense to have a single, shape-shifting robot that can crawl through small holes, climb stairs, cross gaps or go through rubble. The robot's first mission is to save lives. Yim's robo-centipede will allow search-and-rescue personnel to look for survivors in a collapsed building by dropping the robot through a 4-inch hole drilled into the wall.
"When a building collapses from an earthquake or a bomb, etc., then digging for survivors is slow and dangerous for both the rescuers and victims," said Craig Eldershaw, a researcher at Xerox's Palo Alto Research Center, or PARC. "It's difficult to send in a standard robot, since no one knows what it's going to find there. The robot may have narrow pipes to crawl through, rubble to climb over, and some corridors clear of debris where it should try to make good time in moving. But a reconfigurable robot can change shape and adapt to the environment."
Hansel and Gretel, two bots that are being developed by researchers at PARC, use a combination of radio and ultrasound to help them form a map of an uncharted, ruined building as they look for potential survivors. Robin Murphy's team from the Center for Robot-Assisted Search and Rescue, which designed the tank-like Inuktun robots used in the 9/11 rescue efforts, is currently focused on enabling bots to determine their own configurations and choose the form that's safest for the situation. For example, a bot could automatically crouch when going over rubble.
Current demonstrations of the capabilities of shape-shifting robots have mostly focused on changes in locomotion, such as moving multi-legged robots into a snake, rolling them like a tank tread, or going from four legs to two. Daniela Rus' ground robots at the Dartmouth Robotics Lab self-reconfigure into a form that allows them to "flow" over rough ground like water. Other groups have demonstrated changes into useful shapes. For example, small reconfigurable robots could transform themselves into screwdrivers or hammers when needed.
However, most of these transformations have been done under human direction. For the bots to be truly independent, they need to be able to think, an issue that has been the subject of heated debate.
"It's a big question: Should there be a central brain or no brain at all?" said University of Pennsylvania researcher Yim. "There are people who are trying to develop systems which have entirely distributed intelligence -- all the intelligence is distributed over every module. There's another camp which says that you need to have some kind of brain or some kind of hierarchy. For example, if you look at your hand, you have fingers attached to a hand and a hand attached to an arm and multiple arms attached to a body. Similarly you can have one module for the hand which is the brain for all the fingers, and you have another brain at your shoulder which talks to the arms, and so on."
"Let's say you are in a room and there's a door and you are in the shape of a snake," said Yim. "But if you can't go under the door, you have to open the doorknob. The system might recognize all that and say, 'Oh, I need to be a human' in order to open the door. And it theoretically knows how to control the human shape, so then the problem becomes, 'How do I get from the snake to the human shape?'"
Current approaches use a set of known configurations, in effect a toolbox of shapes (a snake, spider or caterpillar, for example). The system figures out the best possible shape for the task and reconfigures the robot from one to another. A more difficult approach, which researchers are still working on, is to look at the task and have the computer automatically generate a shape to do the job.
Pictures:
Source: http://www.wired.com/news/technology/0,1282,65866,00.html?tw=wn_1techhead
"Terminators are coming! Terminators are coming!" :P
