Anybots didn't release any information about what they are doing, but based on the video it is clear that they are doing feedback control. I have no idea why it is so wobbly other than the fact that pneumatic actuators are notoriously hard to model and control and thus some kind of overshoot in the controller is to be expected.
ASIMO and every other successful walking robot has to use feedback control for balancing. Open-loop robots will definitely fall down after only a few steps because unmodeled errors build up quickly.
Honda also did not release all the details of their controller, but they do have a few technical papers in robotics conferences. When ASIMO and other ZMP-based humanoids walk, they use a precomputed "walking pattern" as a reference. The pattern specifies the desired location of the ZMP over time.
There is some limited high-level technical info here:
As the robot walks, the six-axis force/torque sensors in the feet and the gyroscope will allow the robot to estimate the actual ZMP position. The controller will then modify the robot's posture in order to compensate for any errors between the actual and desired ZMP locations. If the controller can keep the error within some reasonable bounds, the trajectory will be dynamically stable and ther robot will not fall down.
By the way, when ASIMO is standing still and the controller is running, if you push on its shoulders or torso, it will resist you. The feedback controller will try to keep the ZMP within the center of the base of support. It moves very smoothly.
ASIMO _absolutely_ dynamically balances. I have one in my lab at CMU and have worked with walking humanoids for years.
First of all, static stability means the center of gravity (CoG) is inside the base of support (the convex hull of the ground contact points).
Dynamic stability is much more difficult to analyze because it involves stability over time. The velocities of any moving parts of an articulated body induce linear and angular momentum that can result in a dynamic stability over time _without_ having any of the intermediate poses being statically stable.
The center of gravity absolutely _does not_ always stay above the support leg for ASIMO. If you try to freeze his pose during the middle of a step it will fall over.
Instead, ASIMO is controlled to keep the Zero Moment Point (ZMP) always inside the convex hull of the contact points. The ZMP (related to the Center of Pressure) is the point on the contact surface where the sum of all torques (moments) is zero. For a given walking trajectory, if the ZMP always stays inside the base if support, then the walk is dynamically stable.
There have been numerous humanoids that use the ZMP formulation to control and maintain dynamic balance while walking (e.g. Honda P2, P3, ASIMO, U. Tokyo H6, H7, AIST HRP2, Waseda Wabian 1, 2, KAIST Hubo, Toyota Partner Robots, Sony QRIO and many more). ALL of these robots are dynamically balancing and are definitely NOT statically stable.
This is a cool robot, but the claim of being the first "dynamically balancing robot" is an overstatement. There have been many dynamically balancing robots before, the most famous being Honda's P2 unveiled in 1997. After that, there have been dozens of walking and dynamically balancing humanoids.
What I think the story _should_ point out that is very impressive is:
1) The robot uses pneumatic actuators, which are notoriously difficult to model and control. Almost all of the current dynamically balancing and walking humanoids use electric motors (e.g. ASIMO).
2) Anybots claims to have some "learning" in their controller. Although they don't have any papers about what they are doing, perhaps they are using some clever statistical modeling and feedback to adaptively control and regulate the robot's stability.
BTW, I had a chance to meet Trevor Blackwell a few years ago when he visited my lab. He is definitely a talented engineer with a vision for the future. Several years ago he made Slashdot when he announced his homemade Segway:
Some of the culturally insensitive postings by slashdot readers is a bit disturbing. Having lived in Japan for several years, it is clear that Western morals regarding suicide do not necessarily apply in this complex and ancient culture.
In addition, readers need to do their homework before posting factual errors or made-up numbers regarding its prevalence. More than 40% of the suicides in Japan are motivated by health-related reasons (older people whose health is failing). These suicide pacts comprise a very small percentage (less than 1%).
Anyway, below are some statistics from 2003 in an article that appeared in The Japan Times on July 23, 2004:
"A record 34,427 people committed suicide in Japan last year. (2003)
The figure, up 7.1 percent from the previous year, remained above 30,000 for the sixth consecutive year, the National Police Agency said in a report released Thursday.
The report says 8,897 people killed themselves over financial difficulties, up 12.1 percent from a year earlier and topping 8,000 for the first time since the NPA began keeping statistics on suicides in 1978.
Suicides motivated by financial difficulties accounted for a quarter of all suicides in the year, comprising the second-largest group, compared with 11.2 percent in 1994.
Almost 60 percent of the suicides in 2003 were by people in their 50s and older, it said.
Health reasons were the motivation for the largest number of suicides in 2003, prompting 15,416, or 44.8 percent of the total, to take their lives. Some 8.5 percent committed suicide due to family problems.
Men accounted for a record 72.5 percent of all suicides in 2003, contributing to the wider gap -- 6.97 years -- between the average life expectancies of men and women, as released earlier this month by the Health, Labor and Welfare Ministry."
A few years ago, I started teaching a game programming
course at Carnegie Mellon. We also had a final project
competition with Xbox and PS2 prizes, as voted by the
students in the class:
It is initially tough to convice some of the older, conservative faculty that
learning how to write games is something that CMU should
be teaching its students. But on second-look, one realizes that
what students really learn is fundamental to all of computer science: efficient data structures, effective resource management and memory usage, good user interfaces, handling images and multimedia content, process threading and multi-user networking, etc. However, with a game programming class, you get to teach all of this stuff in a fun way, where students are extremely self-motivated to learn it all.
The class has been quite popular, and many of my students have gone off
to work in the game development industry. The best feedback I have
received has been from students who enjoyed the fact that their final game projects
have been the the only pieces of software they have written during
their university days that had a lifetime beyond the course itself. I think
game programming is an excellent way to teach coding skills and working as
part of a development team, which is a very practical part of any CS curriculum.
There are downloadable movies of some of the recent lab projects here (all written in portable OpenGL code:
There are many Military Robots that have been actively used in the past. Most of them are drones for mapping or reconnaissance. Note that the Ottawa Treaty of 1999 forbids the production of armed autonomous robots, but South Korea obviously refused to sign the treaty (as did China, Russia, the US, and about 40 others).
The US has used the PackBot in combat situations, but I believe it has never had munitions mounted on it. It looks like iRobot's vacuum cleaning and navigation technology in the Roomba can be reused to "sweep" for land mines in a minefield.
Part of the problem is the default settings of Windows XP Home and Professional. I really wish there was a "secure laptop" Local Security Policy profile that a user could select to automatically configure all of the XP services, etc. Whenever I purchase a new laptop, I have to spend a whole day disabling potentially insecure things like UnPNP, Telnet, Remote Desktop, Remote Registry, SSDP discovery, guest account, default file and printer sharing, etc. and setting up IPSec policies.
What I really want is an easy way to automatically configure these things for a laptop that I NEVER want to be accessed (i.e. remotely controlled) from the outside, nor share any files or resources. It is shameful how many ports are opened by default, which makes the naive user even more prone to picking up nasty trojans and viruses.
IIRC, everyone I remember talking to about it said "the graphics were great, but the plot and characters were shallow and undeveloped."
Yes. The filmmakers spent too much time trying to make a photorealistic film that the story and character development was neglected. And in the end, the audience was unfortunately too busy noticing the realistic light reflections on a character's helmet to pay attention to the storyline anyway...
Over the history of animated films, there has developed the concept of "suspension of disbelief" that allows the audience to forget that they are watching a film and become emotionally attached to the characters. Disney films were particularly good at doing this, despite the fact that the renderings are not "realistic" (Mickey mouse has four fingers, etc.)
For computer animation and video games, audiences are especially critical of mistakes in the depiction of human characters (both appearance and movement), particularly if they attempt to be too "photorealistic". Remember how audiences reacted to the movie "Final Fantasy"? Making characters believable is not simply a function of realistic rendering, but depends deeply on the story and personalities. In the case of Final Fantasy, audiences were too distracted in noticing subtle rendering and animation flaws that they failed to suspend disbelief. If the rendering had been more stylized (like in Japanese anime), viewers may not have focused only on rendering flaws, but on the story and characters. This is what Pixar knows how to do very well, making movies about Toys, Bugs, Cars, etc.
I love the Zelda series, and I love how Nintendo has consistently worked hard at developing storyline and characters, and bringing it all to life with a unique stylized rendering. I simply hope that the new Zelda will not be too "realistic" that players will be distracted by any small flaws in the appearance to appreciate the underlying story.
Actually, the human eye (and optic nervous system) can indeed process events at more than 26 fps. However, the so-called "persistence of vision" prevents the discrimination of distinct events that are too close in time. Biologists have measured the smallest time interval between events to be perceived independently to be roughly 10 milliseconds for the case of a flickering light (which corresponds to 100 Hz).
Note that "frame rate" is different than "flicker rate". Even though movies are filmed at 24 fps, they are actually played back on modern projectors at 48 fps to reduce flicker. This means that you are actually seeing each frame twice. Some newer projectors actually run at 72 fps and display each frame three times. Tests have shown that humans will object to flicker rates lower than 16 as being too distracting. One reason AC current (in the US) runs at 60 Hz is so that we do not see our light bulbs flicker.
Note that smooth motion can be perceived at rates as low as 10 fps, but this depends strongly on the speed of moving objects. Thus, for games with fast-moving objects, the frame rate needs to be higher in order for smooth motion to be perceived. NTSC signals on a TV run at 30 Hz, but are interlaced so that motion appears smoother.
You can find a nice overview of persistence of vision here:
Being a robotics researcher myself, I try to pay attention to news about the latest research. But articles like these are simply painful to read.
I am concerned when these kinds of claims get distributed in the media, because I believe it misleads the public into thinking that the goals of "artificial intelligence" have been achieved (only to have everyone disappointed when things don't actually pan out). This is harmful to robotics research just as a boy "crying wolf" endangers the community.
This professor is famous as one of those researchers who is more interested in getting publicity and attracting media attention than doing serious research. I have seen the "Ubibot" software demo (which I assume the London articles are referring to). To me, there was nothing there that was very interesting. It was something akin to a virtual pet (like those Tamagotchi's from a few years back). There was a very simple model of behaviors (tired, playful, hungry, etc.) that a simple simulated dog was cycling through in a virtual world, and a user could interact with (feed the dog, pet it, etc.).
I have a very hard time believing anyone who claims to have created software that enables robots to feel "emotions" using existing computing hardware. This has more to do with philosophy, semantic definitions, and the fundamental limitations of deterministic machines.
The problem is that this kind of claim is nearly impossible to either prove or disprove because the argument relies on how you define "emotion", "feeling", or "lust". Arguing over semantics is not the way to progress - rather researchers should focus on just getting machines to do useful work. Reproduction is the only way that biological systems can pass along information to future generations. Because it is possible to simply copy the entire contents of a robot's "brain" there is no need for DNA or chromosomes for a robot to pass along information.
This article caught my eye because I am a robotics researcher. Although I admire the intentions of the project, I think such a system design is hard to justify given other alternatives.
Here are some thoughts:
1) Using RFID tags for robot localization is inaccurate and probably not the best choice for fairly structured indoor environments like supermarkets. Just putting colored/patterned tiles on the floor or ceiling, or using indoor GPS would probably allow the robot to navigate better.
2) My gut feeling is that blind people want directions on how to find a particular product. Given simple audio advice similar to how GPS car navigation systems do: "turn left here, walk forward, etc" they would probably be able to use a cane or dog to get to the destination quicker, rather than having to follow around a slow-moving robot. So perhaps it would be better to give blind shoppers a small, portable indoor GPS enabled device when they enter the store that can always tell them where they are and give directions to where they want to go.
3) It would be very difficult to market something like this due to liability concerns. What happens if the robot accidentally leads the person into a sharp object, moving cart, irate grandmother, stairwell, etc... Any injury suffered by anyone involved that gets blamed on the robot will cause the personal injury lawyers to come knocking (especially here in the US, where the power of lawyers is very effective at discouraging new technology from entering society out of fear of liability).
4) As an aside, having RFID tags to identify various products that a blind shopper would want to buy, and then giving them a handheld scanner they could use to check products and prices on shelves might be useful.
Slashdot Mirror
Anybots didn't release any information about what they are doing, but based on the video it is clear that they are doing feedback control. I have no idea why it is so wobbly other than the fact that pneumatic actuators are notoriously hard to model and control and thus some kind of overshoot in the controller is to be expected.
2 .html
ASIMO and every other successful walking robot has to use feedback control for balancing. Open-loop robots will definitely fall down after only a few steps because unmodeled errors build up quickly.
Honda also did not release all the details of their controller, but they do have a few technical papers in robotics conferences. When ASIMO and other ZMP-based humanoids walk, they use a precomputed "walking pattern" as a reference. The pattern specifies the desired location of the ZMP over time.
There is some limited high-level technical info here:
http://asimo.honda.com/
http://world.honda.com/ASIMO/technology/walking_0
As the robot walks, the six-axis force/torque sensors in the feet and the gyroscope will allow the robot to estimate the actual ZMP position. The controller will then modify the robot's posture in order to compensate for any errors between the actual and desired ZMP locations. If the controller can keep the error within some reasonable bounds, the trajectory will be dynamically stable and ther robot will not fall down.
By the way, when ASIMO is standing still and the controller is running, if you push on its shoulders or torso, it will resist you. The feedback controller will try to keep the ZMP within the center of the base of support. It moves very smoothly.
Hope this helps to clarify.
-James
ASIMO _absolutely_ dynamically balances. I have one in my lab at CMU and have worked with walking humanoids for years.
First of all, static stability means the center of gravity (CoG) is inside the base of support (the convex hull of the ground contact points).
Dynamic stability is much more difficult to analyze because it involves stability over time. The velocities of any moving parts of an articulated body induce linear and angular momentum that can result in a dynamic stability over time _without_ having any of the intermediate poses being statically stable.
The center of gravity absolutely _does not_ always stay above the support leg for ASIMO. If you try to freeze his pose during the middle of a step it will fall over.
Instead, ASIMO is controlled to keep the Zero Moment Point (ZMP) always inside the convex hull of the contact points. The ZMP (related to the Center of Pressure) is the point on the contact surface where the sum of all torques (moments) is zero. For a given walking trajectory, if the ZMP always stays inside the base if support, then the walk is dynamically stable.
There have been numerous humanoids that use the ZMP formulation to control and maintain dynamic balance while walking (e.g. Honda P2, P3, ASIMO, U. Tokyo H6, H7, AIST HRP2, Waseda Wabian 1, 2, KAIST Hubo, Toyota Partner Robots, Sony QRIO and many more). ALL of these robots are dynamically balancing and are definitely NOT statically stable.
What I think the story _should_ point out that is very impressive is:
1) The robot uses pneumatic actuators, which are notoriously difficult to model and control. Almost all of the current dynamically balancing and walking humanoids use electric motors (e.g. ASIMO).
2) Anybots claims to have some "learning" in their controller. Although they don't have any papers about what they are doing, perhaps they are using some clever statistical modeling and feedback to adaptively control and regulate the robot's stability.
BTW, I had a chance to meet Trevor Blackwell a few years ago when he visited my lab. He is definitely a talented engineer with a vision for the future. Several years ago he made Slashdot when he announced his homemade Segway:
http://hardware.slashdot.org/article.pl?sid=03/09/ 28/1756241
From my point of view, any interest by hobbyists and industry in humanoid robotics in North America is great for the field of robotics research.
Some of the culturally insensitive postings by slashdot readers is a bit disturbing. Having lived in Japan for several years, it is clear that Western morals regarding suicide do not necessarily apply in this complex and ancient culture.
In addition, readers need to do their homework before posting factual errors or made-up numbers regarding its prevalence. More than 40% of the suicides in Japan are motivated by health-related reasons (older people whose health is failing). These suicide pacts comprise a very small percentage (less than 1%).
Anyway, below are some statistics from 2003 in an article that appeared in The Japan Times on July 23, 2004:
"A record 34,427 people committed suicide in Japan last year. (2003)
The figure, up 7.1 percent from the previous year, remained above 30,000 for the sixth consecutive year, the National Police Agency said in a report released Thursday.
The report says 8,897 people killed themselves over financial difficulties, up 12.1 percent from a year earlier and topping 8,000 for the first time since the NPA began keeping statistics on suicides in 1978.
Suicides motivated by financial difficulties accounted for a quarter of all suicides in the year, comprising the second-largest group, compared with 11.2 percent in 1994.
Almost 60 percent of the suicides in 2003 were by people in their 50s and older, it said.
Health reasons were the motivation for the largest number of suicides in 2003, prompting 15,416, or 44.8 percent of the total, to take their lives. Some 8.5 percent committed suicide due to family problems.
Men accounted for a record 72.5 percent of all suicides in 2003, contributing to the wider gap -- 6.97 years -- between the average life expectancies of men and women, as released earlier this month by the Health, Labor and Welfare Ministry."
http://gamedev.cs.cmu.edu/spring2004/
It is initially tough to convice some of the older, conservative faculty that learning how to write games is something that CMU should be teaching its students. But on second-look, one realizes that what students really learn is fundamental to all of computer science: efficient data structures, effective resource management and memory usage, good user interfaces, handling images and multimedia content, process threading and multi-user networking, etc. However, with a game programming class, you get to teach all of this stuff in a fun way, where students are extremely self-motivated to learn it all.
The class has been quite popular, and many of my students have gone off to work in the game development industry. The best feedback I have received has been from students who enjoyed the fact that their final game projects have been the the only pieces of software they have written during their university days that had a lifetime beyond the course itself. I think game programming is an excellent way to teach coding skills and working as part of a development team, which is a very practical part of any CS curriculum.
There are downloadable movies of some of the recent lab projects here (all written in portable OpenGL code:
http://gamedev.cs.cmu.edu/spring2004/labs/lab1/
http://gamedev.cs.cmu.edu/spring2004/labs/lab2/
There are many Military Robots that have been actively used in the past. Most of them are drones for mapping or reconnaissance. Note that the Ottawa Treaty of 1999 forbids the production of armed autonomous robots, but South Korea obviously refused to sign the treaty (as did China, Russia, the US, and about 40 others).
The US has used the PackBot in combat situations, but I believe it has never had munitions mounted on it. It looks like iRobot's vacuum cleaning and navigation technology in the Roomba can be reused to "sweep" for land mines in a minefield.
Part of the problem is the default settings of Windows XP Home and Professional. I really wish there was a "secure laptop" Local Security Policy profile that a user could select to automatically configure all of the XP services, etc. Whenever I purchase a new laptop, I have to spend a whole day disabling potentially insecure things like UnPNP, Telnet, Remote Desktop, Remote Registry, SSDP discovery, guest account, default file and printer sharing, etc. and setting up IPSec policies.
What I really want is an easy way to automatically configure these things for a laptop that I NEVER want to be accessed (i.e. remotely controlled) from the outside, nor share any files or resources. It is shameful how many ports are opened by default, which makes the naive user even more prone to picking up nasty trojans and viruses.
IIRC, everyone I remember talking to about it said "the graphics were great, but the plot and characters were shallow and undeveloped."
Yes. The filmmakers spent too much time trying to make a photorealistic film that the story and character development was neglected. And in the end, the audience was unfortunately too busy noticing the realistic light reflections on a character's helmet to pay attention to the storyline anyway...
For computer animation and video games, audiences are especially critical of mistakes in the depiction of human characters (both appearance and movement), particularly if they attempt to be too "photorealistic". Remember how audiences reacted to the movie "Final Fantasy"? Making characters believable is not simply a function of realistic rendering, but depends deeply on the story and personalities. In the case of Final Fantasy, audiences were too distracted in noticing subtle rendering and animation flaws that they failed to suspend disbelief. If the rendering had been more stylized (like in Japanese anime), viewers may not have focused only on rendering flaws, but on the story and characters. This is what Pixar knows how to do very well, making movies about Toys, Bugs, Cars, etc.
I love the Zelda series, and I love how Nintendo has consistently worked hard at developing storyline and characters, and bringing it all to life with a unique stylized rendering. I simply hope that the new Zelda will not be too "realistic" that players will be distracted by any small flaws in the appearance to appreciate the underlying story.
Actually, the human eye (and optic nervous system) can indeed process events at more than 26 fps. However, the so-called "persistence of vision" prevents the discrimination of distinct events that are too close in time. Biologists have measured the smallest time interval between events to be perceived independently to be roughly 10 milliseconds for the case of a flickering light (which corresponds to 100 Hz).
n
Note that "frame rate" is different than "flicker rate". Even though movies are filmed at 24 fps, they are actually played back on modern projectors at 48 fps to reduce flicker. This means that you are actually seeing each frame twice. Some newer projectors actually run at 72 fps and display each frame three times. Tests have shown that humans will object to flicker rates lower than 16 as being too distracting. One reason AC current (in the US) runs at 60 Hz is so that we do not see our light bulbs flicker.
Note that smooth motion can be perceived at rates as low as 10 fps, but this depends strongly on the speed of moving objects. Thus, for games with fast-moving objects, the frame rate needs to be higher in order for smooth motion to be perceived. NTSC signals on a TV run at 30 Hz, but are interlaced so that motion appears smoother.
You can find a nice overview of persistence of vision here:
http://en.wikipedia.org/wiki/Persistence_of_visio
Being a robotics researcher myself, I try to pay attention to news about the latest research. But articles like these are simply painful to read.
I am concerned when these kinds of claims get distributed in the media, because I believe it misleads the public into thinking that the goals of "artificial intelligence" have been achieved (only to have everyone disappointed when things don't actually pan out). This is harmful to robotics research just as a boy "crying wolf" endangers the community.
This professor is famous as one of those researchers who is more interested in getting publicity and attracting media attention than doing serious research. I have seen the "Ubibot" software demo (which I assume the London articles are referring to). To me, there was nothing there that was very interesting. It was something akin to a virtual pet (like those Tamagotchi's from a few years back). There was a very simple model of behaviors (tired, playful, hungry, etc.) that a simple simulated dog was cycling through in a virtual world, and a user could interact with (feed the dog, pet it, etc.).
I have a very hard time believing anyone who claims to have created software that enables robots to feel "emotions" using existing computing hardware. This has more to do with philosophy, semantic definitions, and the fundamental limitations of deterministic machines.
The problem is that this kind of claim is nearly impossible to either prove or disprove because the argument relies on how you define "emotion", "feeling", or "lust". Arguing over semantics is not the way to progress - rather researchers should focus on just getting machines to do useful work. Reproduction is the only way that biological systems can pass along information to future generations. Because it is possible to simply copy the entire contents of a robot's "brain" there is no need for DNA or chromosomes for a robot to pass along information.
This article caught my eye because I am a robotics researcher. Although I admire the intentions of the project, I think such a system design is hard to justify given other alternatives.
Here are some thoughts:
1) Using RFID tags for robot localization is inaccurate and probably not the best choice for fairly structured indoor environments like supermarkets. Just putting colored/patterned tiles on the floor or ceiling, or using indoor GPS would probably allow the robot to navigate better.
2) My gut feeling is that blind people want directions on how to find a particular product. Given simple audio advice similar to how GPS car navigation systems do: "turn left here, walk forward, etc" they would probably be able to use a cane or dog to get to the destination quicker, rather than having to follow around a slow-moving robot. So perhaps it would be better to give blind shoppers a small, portable indoor GPS enabled device when they enter the store that can always tell them where they are and give directions to where they want to go.
3) It would be very difficult to market something like this due to liability concerns. What happens if the robot accidentally leads the person into a sharp object, moving cart, irate grandmother, stairwell, etc... Any injury suffered by anyone involved that gets blamed on the robot will cause the personal injury lawyers to come knocking (especially here in the US, where the power of lawyers is very effective at discouraging new technology from entering society out of fear of liability).
4) As an aside, having RFID tags to identify various products that a blind shopper would want to buy, and then giving them a handheld scanner they could use to check products and prices on shelves might be useful.
-James