Wednesday, December 15, 2004
Honda releases an Asimo that jumps!
Honda announced its next-gen Asimo today, probably in response to Toyota's recent unveiling of its "partner robots" and huge i-walk exoskeleton. The new Asimo looks sleeker, and has a much smaller backpack than the earlier models. Its power lasts an hour, compared to 30 minutes for the old model. More important, this robot can jog, in a fashion similar to Sony's QURO. Movies available at Honda's new video website at http://world.honda.com/HDTV/ASIMO/.
Honda's accomplishment is significant because the Asimo is much taller (4 inches more than its predecessor) and more importantly, heavier (about 120 pounds). Since the kinetic energy of a moving object rises at 1/2 mass x velocity squared, the forces that the Asimo must correct for when running are proportionately far greater than for the smaller, QURO-sized robots.
To allow the Asimo to jog, the new Asimo can bend and twist its torso relative to its legs. This is accomplished by a new hip joint on the body. The HRP-2 humanoid developed by General Robotix has a hip joint, but in general other robots do not, accounting for some of their stiffness when walking. The hip joint allows the Asimo to run at 3km/hr, and allows faster, more natural-looking walking at about 2km/hr (which means that the new Asimo could go up to 3 miles between charges). In general, one would expect that Asimo's movements would look less machine-like, since humans constantly readjust their hip/spine during movement.
Asimo also has to be stronger to run. Running puts far greater loads on its joints and motors, which must also have some "give" to avoid bending and breaking. According to Honda, the new Asimo has more powerful motors and a special (pattern generator?) circuit that repetitively cycles the leg motion. It is not clear how shock is absorbed by the "lightweight" leg assembly - do the new motors have elastic give like those created by US company Yobotics?
In addition to running, Asimo is now a bit smarter about moving. Previous versions pre-planned their movement, and then inflexibly executed according to a model of the environment. The new model dynamically adjusts progress based on feedback from the foot sensors and obstacles detected by machine vision.
It is not clear if this is separate from Asimo's previous ability to notice moving objects (e.g. human walkers) crossing its path, and adjusting so it doesn't run into them.
Additional force sensors have been put into the hands, and the thumb can now bend independently of the fingers. This allows Asimo to determine if, for example, it hand is being shaken, and dynamically adjust the tension in its arm. Previous models apparently let the arm go limp during a handshake and did not sense the shake itself. The new model can detect if its arm is being pushed or pulled, and move accordingly. This opens the possibility of an Asimo "dancing" with a human partner, and taking force cues from its dance partner as to who is "leading" or "following".
Finally, there are more joints. In addition to the hip rotation joint, the wrist can now bend in three axes, and the thumb has a motor allowing independent operation. The neck joint now has a second axis, allows more "expressiveness" - in other words, a head tilted relative to the torso.
Overall, Honda has kept itself in the running, even though the new model does not clearly blow away the Toyota and especially the Sony QURO robots. Honda is following an incremental strategy in developing the Asimo into a practical consumer product.
However, we are still late in 2004 - I wouldn't be surprised to see more advances in the Asimo by mid-2005, when the EXPO 2005 in Aichi, Japan opens its "robot pavilion."
Honda announced its next-gen Asimo today, probably in response to Toyota's recent unveiling of its "partner robots" and huge i-walk exoskeleton. The new Asimo looks sleeker, and has a much smaller backpack than the earlier models. Its power lasts an hour, compared to 30 minutes for the old model. More important, this robot can jog, in a fashion similar to Sony's QURO. Movies available at Honda's new video website at http://world.honda.com/HDTV/ASIMO/.
Honda's accomplishment is significant because the Asimo is much taller (4 inches more than its predecessor) and more importantly, heavier (about 120 pounds). Since the kinetic energy of a moving object rises at 1/2 mass x velocity squared, the forces that the Asimo must correct for when running are proportionately far greater than for the smaller, QURO-sized robots.
To allow the Asimo to jog, the new Asimo can bend and twist its torso relative to its legs. This is accomplished by a new hip joint on the body. The HRP-2 humanoid developed by General Robotix has a hip joint, but in general other robots do not, accounting for some of their stiffness when walking. The hip joint allows the Asimo to run at 3km/hr, and allows faster, more natural-looking walking at about 2km/hr (which means that the new Asimo could go up to 3 miles between charges). In general, one would expect that Asimo's movements would look less machine-like, since humans constantly readjust their hip/spine during movement.
Asimo also has to be stronger to run. Running puts far greater loads on its joints and motors, which must also have some "give" to avoid bending and breaking. According to Honda, the new Asimo has more powerful motors and a special (pattern generator?) circuit that repetitively cycles the leg motion. It is not clear how shock is absorbed by the "lightweight" leg assembly - do the new motors have elastic give like those created by US company Yobotics?
In addition to running, Asimo is now a bit smarter about moving. Previous versions pre-planned their movement, and then inflexibly executed according to a model of the environment. The new model dynamically adjusts progress based on feedback from the foot sensors and obstacles detected by machine vision.
It is not clear if this is separate from Asimo's previous ability to notice moving objects (e.g. human walkers) crossing its path, and adjusting so it doesn't run into them.
Additional force sensors have been put into the hands, and the thumb can now bend independently of the fingers. This allows Asimo to determine if, for example, it hand is being shaken, and dynamically adjust the tension in its arm. Previous models apparently let the arm go limp during a handshake and did not sense the shake itself. The new model can detect if its arm is being pushed or pulled, and move accordingly. This opens the possibility of an Asimo "dancing" with a human partner, and taking force cues from its dance partner as to who is "leading" or "following".
Finally, there are more joints. In addition to the hip rotation joint, the wrist can now bend in three axes, and the thumb has a motor allowing independent operation. The neck joint now has a second axis, allows more "expressiveness" - in other words, a head tilted relative to the torso.
Overall, Honda has kept itself in the running, even though the new model does not clearly blow away the Toyota and especially the Sony QURO robots. Honda is following an incremental strategy in developing the Asimo into a practical consumer product.
However, we are still late in 2004 - I wouldn't be surprised to see more advances in the Asimo by mid-2005, when the EXPO 2005 in Aichi, Japan opens its "robot pavilion."
Tuesday, December 14, 2004
Roboburgh, Centaurs, Robot fish and living machines
One of the most interesting questions about the future of robotics is the location of the equivalent of the robotic equivalent of Silicon Valley - a "robot valley" which will be the world center of robotics development. To date, the city of Osaka, Japan and Pittsburgh seem to be top contenders. A key feature of any emerging robo-corridor is the presence of business associations and startup incubators fostering a commercial robotics presence.
This week, Pittsburgh made a move to strengthen its robo-corridor by merging Pittsburgh Digital Greenhouse and the Robotics Foundry. Apparently, both groups have been floundering a bit since their inception a few years ago. The merger will allow both groups to pool resources. The Greenhouse lists its efforts as helping a group of companies develop advanced network technologies, and a 'system on a chip' (don't we have that already?). The Robotics Foundry tries to promote robotics-related companies in the Pittsburgh area. On Dec. 15th, it announced a major score - $27 million government contract to research and develop defense-related robotic technology. According to an article in the Pittsburgh Business Journal:
"The Robotics Foundry has been struggling over the last year or two to really get things going in Pittsburgh and I think their ability to shine a spotlight on the robotics industry has led to government contract dollars coming into Western Pennsylvania."
It is not obvious that the two missions match. Networking chips are only marginally useful for robotics, and a 'system on a chip' is interesting only to the extent that it allows multiple cpus to govern different portions of a single robot. As I have commented before, robotic and cyberspace computing are not equivalent - instead they are mirror images of each other. PCs and networks try to create an artificial world that people enter via a (usually graphical) interface. In contrast, robots strive to enter our world via real-time processing of environmental data. The computing requirements are vastly different.
Even so, a merger of these two groups may signal some strengthening of the Pittsburgh bid to be "robo-valley."
Today seems to be the day for robotic critters. First off is a robotic fish created by Chinese researchers. This black-banded fish-bot is about a meter long and can swim horizontally and vertically using a flexible tail instead of propellers. According to various Internet articles, the robot will be used for underwater archeology, though uses in ocean environmental monitoring and even spying seem quite likely.
Clearly, the concept of biomorphic robots is interesting to Chinese researcher. To me, it is fascinating that China is clearly working very hard at advanced robotics according to the "Asian" model - animal-like and humanoid systems for operation in normal environments. In contrast, much of the US academic and government/defense robotic industry clings to the notion of special-purpose machines which can only function in ultra-specialized environments without humans.
Another biomorph recently released is in my favorite form - a centaur. This one comes from Helsinki University's WorkPartner Project, with a website at http://www.automation.hut.fi/IMSRI/workpartner/index.html. This system is on four wheels, but has a humanoid torso similar to NASA's Robonaut.
Looking at these robots - which emphasize body over any kind of AI "mind" - brings up the question as to what extent these machines are "living" rather than "intelligent". Here are some common features used to describe the living state, and the status of mobile robots for each feature:
1. Energy release used to create structure ("negative entropy") - Here, robots clearly are one with living things - but also with other machines, clouds, volcanoes, and other physical entitites that create and maintain their structure via a flow of energy. For an example of a 'plantlike' robot, see the description of recent underwater robots which have solar cells to recharge themselves. Interestingly, one company, Husqvarna, has created a solar-powered 'mobot' Automower for large lawns. In the future, robots will probably use fuel cells, which will allow them to 'eat' certain kinds of materials, even food like starch, to produce energy.
2. Growth and differentiation - Most robots don't show this, the pseudo-growth of the Aibo ERS robo-dog nonwithstanding. Real behavioral growth would allow unprogrammed changes in behavior that conferred long-term benefit. With respect to physical growth, recent experiments at Xerox PARC with assembling modular robots from lots of interlocking building blocks appear to provide this feature.
3. Reproduction - To date, robots do not reproduce. To reproduce, a robot would need to direct the assembly of another robot. This seems likely in a few years.
4. Genetic/heredity - Note that this is NOT the same as reproduction. To mimic the genetic system of living things (and thereby allow for the potential for evolutionary change), a robot would have to direct the assembly of an "imperfect copy" of itself. Furthermore, the newly created robot would need to be able to "pass on" these imperfections when it makes a copy of itself. This would qualify as clonal, or asexual heredity. If robots could mix the data describing their construction together, drawn from several individuals, it would qualify as sexual reproduction. Some timid steps are being made in this direction with work at Brandeis university (Lamarkian heredity) and a recent report from Korea about robots implementing chromosome analogs in software.
One of the most interesting questions about the future of robotics is the location of the equivalent of the robotic equivalent of Silicon Valley - a "robot valley" which will be the world center of robotics development. To date, the city of Osaka, Japan and Pittsburgh seem to be top contenders. A key feature of any emerging robo-corridor is the presence of business associations and startup incubators fostering a commercial robotics presence.
This week, Pittsburgh made a move to strengthen its robo-corridor by merging Pittsburgh Digital Greenhouse and the Robotics Foundry. Apparently, both groups have been floundering a bit since their inception a few years ago. The merger will allow both groups to pool resources. The Greenhouse lists its efforts as helping a group of companies develop advanced network technologies, and a 'system on a chip' (don't we have that already?). The Robotics Foundry tries to promote robotics-related companies in the Pittsburgh area. On Dec. 15th, it announced a major score - $27 million government contract to research and develop defense-related robotic technology. According to an article in the Pittsburgh Business Journal:
"The Robotics Foundry has been struggling over the last year or two to really get things going in Pittsburgh and I think their ability to shine a spotlight on the robotics industry has led to government contract dollars coming into Western Pennsylvania."
It is not obvious that the two missions match. Networking chips are only marginally useful for robotics, and a 'system on a chip' is interesting only to the extent that it allows multiple cpus to govern different portions of a single robot. As I have commented before, robotic and cyberspace computing are not equivalent - instead they are mirror images of each other. PCs and networks try to create an artificial world that people enter via a (usually graphical) interface. In contrast, robots strive to enter our world via real-time processing of environmental data. The computing requirements are vastly different.
Even so, a merger of these two groups may signal some strengthening of the Pittsburgh bid to be "robo-valley."
Today seems to be the day for robotic critters. First off is a robotic fish created by Chinese researchers. This black-banded fish-bot is about a meter long and can swim horizontally and vertically using a flexible tail instead of propellers. According to various Internet articles, the robot will be used for underwater archeology, though uses in ocean environmental monitoring and even spying seem quite likely.
Clearly, the concept of biomorphic robots is interesting to Chinese researcher. To me, it is fascinating that China is clearly working very hard at advanced robotics according to the "Asian" model - animal-like and humanoid systems for operation in normal environments. In contrast, much of the US academic and government/defense robotic industry clings to the notion of special-purpose machines which can only function in ultra-specialized environments without humans.
Another biomorph recently released is in my favorite form - a centaur. This one comes from Helsinki University's WorkPartner Project, with a website at http://www.automation.hut.fi/IMSRI/workpartner/index.html. This system is on four wheels, but has a humanoid torso similar to NASA's Robonaut.
Looking at these robots - which emphasize body over any kind of AI "mind" - brings up the question as to what extent these machines are "living" rather than "intelligent". Here are some common features used to describe the living state, and the status of mobile robots for each feature:
1. Energy release used to create structure ("negative entropy") - Here, robots clearly are one with living things - but also with other machines, clouds, volcanoes, and other physical entitites that create and maintain their structure via a flow of energy. For an example of a 'plantlike' robot, see the description of recent underwater robots which have solar cells to recharge themselves. Interestingly, one company, Husqvarna, has created a solar-powered 'mobot' Automower for large lawns. In the future, robots will probably use fuel cells, which will allow them to 'eat' certain kinds of materials, even food like starch, to produce energy.
2. Growth and differentiation - Most robots don't show this, the pseudo-growth of the Aibo ERS robo-dog nonwithstanding. Real behavioral growth would allow unprogrammed changes in behavior that conferred long-term benefit. With respect to physical growth, recent experiments at Xerox PARC with assembling modular robots from lots of interlocking building blocks appear to provide this feature.
3. Reproduction - To date, robots do not reproduce. To reproduce, a robot would need to direct the assembly of another robot. This seems likely in a few years.
4. Genetic/heredity - Note that this is NOT the same as reproduction. To mimic the genetic system of living things (and thereby allow for the potential for evolutionary change), a robot would have to direct the assembly of an "imperfect copy" of itself. Furthermore, the newly created robot would need to be able to "pass on" these imperfections when it makes a copy of itself. This would qualify as clonal, or asexual heredity. If robots could mix the data describing their construction together, drawn from several individuals, it would qualify as sexual reproduction. Some timid steps are being made in this direction with work at Brandeis university (Lamarkian heredity) and a recent report from Korea about robots implementing chromosome analogs in software.
