Tuesday, January 04, 2005
Robots that jump in 2005
Here are a few predictions for 2005 concerning 'new era' robotics - that is, robots designed to work in real-world environments (instead of a lab project designed to get a degree) with emphasis on body, versus "mind" development.
1. The 1990s PC/Internet vision of the future will show major cracks. For the past several years, we've seen 1990s ideas recycled as 'new' - the ever-present 24/7 prediction for constant access to the 'Net, wearable computers, wireless everywhere, more game polygons, etc. In 2005 parts of the technology industry will realize that these aren't new tech - that instead they're still chasing the last revolution. True, these technologies will expand, but that expansion will be leveling out - at the upper end of a hyperbolic curve - as the PC/Internet industry continues its march to power company status. After the power coming out of your wall socket is amazing - but how many people see it as 'hot tech'? Increasingly, well see webloggers and even a few 'tech pundits' noticing that we aren't driving into a new era - we're simply filling up the corners of an already complete tech revolution. An a very few will start looking elsewhere for the 'next big thing.'
2. Robotics will continue is move from Hollywood fantasy to a new, real-world vision. For decades, it is been ok for even a highly sophisticated PC developer or Internet guru to make a moronic prediction that "robots will take over and enslave us." Instead of a serious consideration of robotics, it was just fine for an otherwise tech-savvy expert to mouth a line from a Hollywood movie when robotics came up. This is because (see next prediction) the typical PC/Internet guru doesn't know anything more about robots that a janitor. In 2005, recycling movie plotline will not be seen as 'clever' - instead, increasing numbers of people will want the real story on robotics. Sadly, most "tech experts" won't be able to tell thim a thing, since they, at best, think robotics is a branch of their industry - just like the stereo/hi-fi industry probably thought personal computers were a side-branch of their industry in 1979.
3. There will be a growing realization that robots are not PCs with wheels, and understanding PCs does not make you an expert on robots. One of the problems the typical tech guru is having with robotics is that their own industry blinds them to its uniqueness. Robots use computer boards, memory, power supplies, processors - all the stuff you might find in a PC. Also, many robots (e.g., Sony's Aibo) can execute PC like functions like checking email. Because of this, many people still see robotics as a minor branch of the big PC/Internet world.
Of course this is not true. PC developers in the 1970s hijacked components from the electronic industry, but the personal computer is not a side branch of 1970s electronics. In the same way, robots are currently using parts hijacked from other industries, but the thing being made is not a PC on wheels. Even "PC robots" like White Box are not really PCs - they use PC technology to do something completely different from the standard PC.
The key difference between PCs, networks, and robots how the computer interacts. In the PC model, the computer makes a little toy world called 'cyberspace' which accepts only symbolic information. People have to learn the right sequence of symbols (commands) to control this toy world. In addition, they must adjust the parameters of the toy world (configuration). They then enter it, completely at the mercy of the machjine's rules. It is the responsibility of the human to interpret the PC's toy cyberspace world, and act in it according to the rules.
In contrast, robots are sensor-laden, and devote most of their processing power to either sampling non-symbolic, environmental information, or executing commands in the real world. In effect, it is the responsibility of the robot to interpret our world, and act in it. Unless you really think reality is a computer simulation, this makes robots and PCs polar opposites, even though they share the same hardware.
During 2005, many software and hardware developers will realize for the first time that programming like Linux and Adobe Acrobat has nothing to do with robots. These are programs which either create or act within a virtual cyberspace. In contrast, robotic programming typically uses a "behavior-based" approach very different from PC or network programming. This is because the robot must interpret and process unreliable sensory data in real time, and act in a real world where you can't "reboot" if you fall off a cliff. Programmers experience in standard programming methods will begin to explore robots, and rapidly realize that they don't know anything. Most will return to cyberspace, but increasing numbers will begin 'robotic programming' - and increasingly define robots as something completely different from a PC.
4. 2005 is the year of giant robots
Thus far, humanoid robots have been built small (e.g. Sony and various small commercial bipeds) and medium (Asimo, HRP). I predict that 2005 will be the year that experiments begin with much larger humanoids. These will be large enough that a person can sit within them, so the robot will also act as an exoskeleton. Examples have already appeared. On the small end, we have Toyota's exoskeleton - bigger than a person, though designed for a relatively small 'pilot.' We also have TMUSK 11-foot high tele-operated robot called the Enryu for rescue missions- also more exoskeleton than autonomous robot. Finally, the quirky startup Neogeotronix is working on a 20-foot humanoid with a design matching classic Japanese animation.
5. Hollywood will continue to be oblivious to real-world robotics
While real-world robots mark a long-term cultural trend away from so-called "cyberspace" and a "turn to the real" in society, Hollywood will continue pushing old dogs like "media convergence", "Internet TV", and various attacks against file-trading. Robotic machines will appear in two very likely to be lame animated movies - Robots and Cars. The topics of these films seem to say that many people believe that computer animation alone is enough to make a film interesting - but more important, neither film will have the smallest cultural connection to the rise or real robots and robotic cars. They are likely to be far worse than the passable if predictable action in I, Robot last year. But this train is coming - ultimately, who would want to watch a robot onscreen when they can have one for real to play with? It may be that WowWee toys will be the next "media giant" with Robosapien going forth to multiply during the next decade - only this time the media will be physical, and robotic.
6. Design will become increasingly important in robots
For decades, robots have "lived" in research labs and government contractor buildings with little connection to the real world or public. Because of the culture of both institutions, design and style have never played much of a role in how robots look and move - hence the bundle of wires lurching across a lab floor or the clumsy trash on treads can being demonstrated for the military. However, with several companies making robots using teams of engineers and designers, it is becoming harder and harder to have a bag of bolts taken seriously. Increasingly, robot designers will focus on the visual, behavioral, and movement design fo their creations, seeing the artistic as well as the purely engineering side of things. It might even lead to a new breed of robot designers uniting science and art - long true, but banished in the 20th century. Increasingly, we'll see fewer piles of wires, and more integrated systems like those at Robo-Garage.
7. Robonexus will be joined by a second 'new era' robot tradeshow in the US
The success of Robonexus at attracting an audience will make other groups think: is this the beginning of something big? After all, the big personal computing tradeshows were once small. Robonexus is likely to be held on the west coast, but other shows - bigger than typical hobby shows, but smaller than full-grown tradeshows like Macworld - will begin to appear. Some may specialize in new-era robots of specific kinds - service, security, healthcare. We should also see greater media coverage of these shows, and prehaps even more visits for advanced Asian humanoics.
8. Nobody will win the 2005 Grand Challenge, but it will be much closer
In 2004, the furthest any of the driverless cars got in the DARPA Grand Challenge was about 7 miles - tiny compared to the 150 mile courseway. In 2005 it seems unlikely that anyone will make it - but I wouldn't be surprised if more than one team got halfway on the course. Compared to 2004, the returning teams will have had years, instead of months to get their robots going. The visibility of the Challenge has spurred driverless car development efforts worldwide, and many independent projects (e.g. visual recognition of roadsigns) have not become available on the Internet. One more prediction: despite two vehicles, Highlander and Sandstorm, I doubt that CMU will win, or even be "number one." Instead, look for interesting upstars like Digital Auto Drive or Cyberrider to go the furthest.
9. A Moore's Law for sensors and robot computation
In 2004, Intel announced that it was not going to make a 4GHz Pentium in the forseeable future. This signals at least a several year pause in the rapid advancement of Moore's Law for computing. However, for robots, more people will realize that sheer computing speed is not the most critical feature of robotic advancement. Instead, sensors will take central stage, and an equivalent of Moore's Law will be defined for robots based on sensor density. The correlation between more sensors and better robotic performance will be recognized, and the emergence of practical humanoids will not be defined in terms of computing power but fine-grainedness of sensors.
Of course, sensors require increasing processing as their numbers increase. In the past, when everything went through a single CPU, more sensors meant more speed was required. But the new direction of the microprocessor industry into "multicore" will change this. Multicore processors combine two or more complete CPUs into a single system, and provide tools for handling computing threads in an efficient manner. While attention is currently focused on 2 and 4 core systems for workstations and servers, robots will benefit from lower-powered processors with 32, 62, 128, and more cores. These multicore systems are ideal for handling increasing sensor loads, and hierarchies of multicore systems feeding upwards in a tree fashion come close to mimicing many aspects of animal nervous systems. True, total computing is going up, but since it is collection of largely independent tasks on distinct cores it really isn't about brute speed like Moore's law. Another way to look at it - each time the number of cores doubles, robots will double in performance, while PCs and network applications will get much smaller speed increases.
10. 3D printing will be used to design robot bodies
Currently, many CAD/CAM systems print not to paper but to 3-D printers capable of creating objects in metal or plastic. As 3D printers get the ability to "print" more than one material in an object, they will be ideal for robotic design. One of the major limitations of robotics has been the confinement to "trash can on wheels" designs necessitated by using cast-off parts from other industries. I expect that at least one group will display a robot designed completely in a CAD/CAM system, then "printed" partly or completely. This will allow the greater design freedom needed by any robot that jumps.
Here are a few predictions for 2005 concerning 'new era' robotics - that is, robots designed to work in real-world environments (instead of a lab project designed to get a degree) with emphasis on body, versus "mind" development.
1. The 1990s PC/Internet vision of the future will show major cracks. For the past several years, we've seen 1990s ideas recycled as 'new' - the ever-present 24/7 prediction for constant access to the 'Net, wearable computers, wireless everywhere, more game polygons, etc. In 2005 parts of the technology industry will realize that these aren't new tech - that instead they're still chasing the last revolution. True, these technologies will expand, but that expansion will be leveling out - at the upper end of a hyperbolic curve - as the PC/Internet industry continues its march to power company status. After the power coming out of your wall socket is amazing - but how many people see it as 'hot tech'? Increasingly, well see webloggers and even a few 'tech pundits' noticing that we aren't driving into a new era - we're simply filling up the corners of an already complete tech revolution. An a very few will start looking elsewhere for the 'next big thing.'
2. Robotics will continue is move from Hollywood fantasy to a new, real-world vision. For decades, it is been ok for even a highly sophisticated PC developer or Internet guru to make a moronic prediction that "robots will take over and enslave us." Instead of a serious consideration of robotics, it was just fine for an otherwise tech-savvy expert to mouth a line from a Hollywood movie when robotics came up. This is because (see next prediction) the typical PC/Internet guru doesn't know anything more about robots that a janitor. In 2005, recycling movie plotline will not be seen as 'clever' - instead, increasing numbers of people will want the real story on robotics. Sadly, most "tech experts" won't be able to tell thim a thing, since they, at best, think robotics is a branch of their industry - just like the stereo/hi-fi industry probably thought personal computers were a side-branch of their industry in 1979.
3. There will be a growing realization that robots are not PCs with wheels, and understanding PCs does not make you an expert on robots. One of the problems the typical tech guru is having with robotics is that their own industry blinds them to its uniqueness. Robots use computer boards, memory, power supplies, processors - all the stuff you might find in a PC. Also, many robots (e.g., Sony's Aibo) can execute PC like functions like checking email. Because of this, many people still see robotics as a minor branch of the big PC/Internet world.
Of course this is not true. PC developers in the 1970s hijacked components from the electronic industry, but the personal computer is not a side branch of 1970s electronics. In the same way, robots are currently using parts hijacked from other industries, but the thing being made is not a PC on wheels. Even "PC robots" like White Box are not really PCs - they use PC technology to do something completely different from the standard PC.
The key difference between PCs, networks, and robots how the computer interacts. In the PC model, the computer makes a little toy world called 'cyberspace' which accepts only symbolic information. People have to learn the right sequence of symbols (commands) to control this toy world. In addition, they must adjust the parameters of the toy world (configuration). They then enter it, completely at the mercy of the machjine's rules. It is the responsibility of the human to interpret the PC's toy cyberspace world, and act in it according to the rules.
In contrast, robots are sensor-laden, and devote most of their processing power to either sampling non-symbolic, environmental information, or executing commands in the real world. In effect, it is the responsibility of the robot to interpret our world, and act in it. Unless you really think reality is a computer simulation, this makes robots and PCs polar opposites, even though they share the same hardware.
During 2005, many software and hardware developers will realize for the first time that programming like Linux and Adobe Acrobat has nothing to do with robots. These are programs which either create or act within a virtual cyberspace. In contrast, robotic programming typically uses a "behavior-based" approach very different from PC or network programming. This is because the robot must interpret and process unreliable sensory data in real time, and act in a real world where you can't "reboot" if you fall off a cliff. Programmers experience in standard programming methods will begin to explore robots, and rapidly realize that they don't know anything. Most will return to cyberspace, but increasing numbers will begin 'robotic programming' - and increasingly define robots as something completely different from a PC.
4. 2005 is the year of giant robots
Thus far, humanoid robots have been built small (e.g. Sony and various small commercial bipeds) and medium (Asimo, HRP). I predict that 2005 will be the year that experiments begin with much larger humanoids. These will be large enough that a person can sit within them, so the robot will also act as an exoskeleton. Examples have already appeared. On the small end, we have Toyota's exoskeleton - bigger than a person, though designed for a relatively small 'pilot.' We also have TMUSK 11-foot high tele-operated robot called the Enryu for rescue missions- also more exoskeleton than autonomous robot. Finally, the quirky startup Neogeotronix is working on a 20-foot humanoid with a design matching classic Japanese animation.
5. Hollywood will continue to be oblivious to real-world robotics
While real-world robots mark a long-term cultural trend away from so-called "cyberspace" and a "turn to the real" in society, Hollywood will continue pushing old dogs like "media convergence", "Internet TV", and various attacks against file-trading. Robotic machines will appear in two very likely to be lame animated movies - Robots and Cars. The topics of these films seem to say that many people believe that computer animation alone is enough to make a film interesting - but more important, neither film will have the smallest cultural connection to the rise or real robots and robotic cars. They are likely to be far worse than the passable if predictable action in I, Robot last year. But this train is coming - ultimately, who would want to watch a robot onscreen when they can have one for real to play with? It may be that WowWee toys will be the next "media giant" with Robosapien going forth to multiply during the next decade - only this time the media will be physical, and robotic.
6. Design will become increasingly important in robots
For decades, robots have "lived" in research labs and government contractor buildings with little connection to the real world or public. Because of the culture of both institutions, design and style have never played much of a role in how robots look and move - hence the bundle of wires lurching across a lab floor or the clumsy trash on treads can being demonstrated for the military. However, with several companies making robots using teams of engineers and designers, it is becoming harder and harder to have a bag of bolts taken seriously. Increasingly, robot designers will focus on the visual, behavioral, and movement design fo their creations, seeing the artistic as well as the purely engineering side of things. It might even lead to a new breed of robot designers uniting science and art - long true, but banished in the 20th century. Increasingly, we'll see fewer piles of wires, and more integrated systems like those at Robo-Garage.
7. Robonexus will be joined by a second 'new era' robot tradeshow in the US
The success of Robonexus at attracting an audience will make other groups think: is this the beginning of something big? After all, the big personal computing tradeshows were once small. Robonexus is likely to be held on the west coast, but other shows - bigger than typical hobby shows, but smaller than full-grown tradeshows like Macworld - will begin to appear. Some may specialize in new-era robots of specific kinds - service, security, healthcare. We should also see greater media coverage of these shows, and prehaps even more visits for advanced Asian humanoics.
8. Nobody will win the 2005 Grand Challenge, but it will be much closer
In 2004, the furthest any of the driverless cars got in the DARPA Grand Challenge was about 7 miles - tiny compared to the 150 mile courseway. In 2005 it seems unlikely that anyone will make it - but I wouldn't be surprised if more than one team got halfway on the course. Compared to 2004, the returning teams will have had years, instead of months to get their robots going. The visibility of the Challenge has spurred driverless car development efforts worldwide, and many independent projects (e.g. visual recognition of roadsigns) have not become available on the Internet. One more prediction: despite two vehicles, Highlander and Sandstorm, I doubt that CMU will win, or even be "number one." Instead, look for interesting upstars like Digital Auto Drive or Cyberrider to go the furthest.
9. A Moore's Law for sensors and robot computation
In 2004, Intel announced that it was not going to make a 4GHz Pentium in the forseeable future. This signals at least a several year pause in the rapid advancement of Moore's Law for computing. However, for robots, more people will realize that sheer computing speed is not the most critical feature of robotic advancement. Instead, sensors will take central stage, and an equivalent of Moore's Law will be defined for robots based on sensor density. The correlation between more sensors and better robotic performance will be recognized, and the emergence of practical humanoids will not be defined in terms of computing power but fine-grainedness of sensors.
Of course, sensors require increasing processing as their numbers increase. In the past, when everything went through a single CPU, more sensors meant more speed was required. But the new direction of the microprocessor industry into "multicore" will change this. Multicore processors combine two or more complete CPUs into a single system, and provide tools for handling computing threads in an efficient manner. While attention is currently focused on 2 and 4 core systems for workstations and servers, robots will benefit from lower-powered processors with 32, 62, 128, and more cores. These multicore systems are ideal for handling increasing sensor loads, and hierarchies of multicore systems feeding upwards in a tree fashion come close to mimicing many aspects of animal nervous systems. True, total computing is going up, but since it is collection of largely independent tasks on distinct cores it really isn't about brute speed like Moore's law. Another way to look at it - each time the number of cores doubles, robots will double in performance, while PCs and network applications will get much smaller speed increases.
10. 3D printing will be used to design robot bodies
Currently, many CAD/CAM systems print not to paper but to 3-D printers capable of creating objects in metal or plastic. As 3D printers get the ability to "print" more than one material in an object, they will be ideal for robotic design. One of the major limitations of robotics has been the confinement to "trash can on wheels" designs necessitated by using cast-off parts from other industries. I expect that at least one group will display a robot designed completely in a CAD/CAM system, then "printed" partly or completely. This will allow the greater design freedom needed by any robot that jumps.
