[2001-05-11]
RoboCup is a robotic soccer contest which was started in 199? to promote robotics research. The organizers of RoboCup have stated a goal for the RoboCup tournament: humanoid robots which could beat the men's World Cup soccer team by 2050.
Recently this goal was discussed at length in the Robot Papers Group of the Dartmouth Robotics Laboratory (the RPG is a weekly meeting of members of the DRL to discuss robotics papers and topics of interest). The remainder of this article presents my thoughts on the 2050 goal.
The RoboCup 2050 Goal is a bold challenge to the robotics community. At the best it is very ambitious, at the worst is is foolhardy. There two questions that can be posed about stating a goal such as this:
1) Is it achievable, and
2) Based on #1 is it a worthwhile goal.
Starting with #2, let's suppose it is an achievable goal. Then the question of its worth is reduced to an analysis of whether the time & effort required to achieve the goal is justified. A classic example of this is John F. Kennedy's 1961 goal of putting a man on the moon before 1970. In hindsight this goal was achievable, although many scientists did not believe it was possible when it was proposed. Given the magnitude of the project, is is easy to understand why many were skeptical. However, even though it was realistic was it worthwhile?
The motive for putting a man on the moon was not scientific. It was motivated by the Cold War existing between the USSR and the USA. When the USSR put Sputnik in orbit in the late 1950's it sent shock waves throughout the United States. People were asking how the Soviet Union could be ahead of the US in rocket research. The military was concerned that this could alter the balance of power in the world. A series of rocket explosions which occured as the US tried to play catch-up further fueled the fires, so to speak. By the time JFK took office the citizens of the USA were losing faith in their technological superiority over the USSR. JFK need a way to challenge the people of the US to regain the lead in space, and the goal of "landing a man on the moon and returning him safely to Earth" was his daring response. However, in retrospect was this worthwhile? Certainly many new technologies were developed, new materials were developed, the electronics industry made great advances, and many jobs were created. But on the other hand an incredible amount of money was spent to go to a place that had no practical value. True, the US did regain the lead in space and everyone on Earth acknowledges it as one of Mankind's greatest achievements, but the public quickly lost interest and the Apollo program was curtailed. Furthermore, we've never gone back and we have no plans to return.
The point is that even achievable goals need to be examined to see if they are worthwhile--especially when they are publically funded. What if the money spent on the Space Race had been spent on medical research or on developing better farming methods. Or even on feeding everyone in the US (yes, while billions were spent putting a man on the moon people were going hungry in America). Would using that money in a different way have made things better? That's always hard to say, but it's a question that deserves to be asked.
As for RoboCup, would the expense of developing a team of robots to defeat the World Cup champion soccer team in 2050 be worthwhile? In order to say "yes" you would have to show that the technology developed in the effort would be of benefit to Mankind--it is unlikely that people would say the accomplishment in itself would justify the expense (does anyone other than a robotocist think it's useful?; does anyone want to watch robots play soccer against humans? -- maybe so until the humans lose, then who cares?). So how useful would the technology be? Certainly we would be able to build very human-like robots with that technology, and these robots could be used to perform dangerous tasks (fight fires, rescue people, work in space) and do labor- intensive tasks we don't like doing. But is RoboCup the only way this technology could be invented? Wouldn't this technology be developed anyway, although possibly at a slower rate? And what is the cost of developing these "super-soccer" robots?
As with most things, it all comes down to money in the end. Should large sums of money be spent to meet the RoboCup 2050 goal? In my opinion, no. If scientists want to spend their normal research funds to try and meet the goal, okay, but I don't think the NSF should be funding multi-million dollar grants to make soccer-playing robots. I want to be clear that I'm not anti-RoboCup. People enjoy the tournaments and the publicity is good for robotics. I'm just saying that developing a World Cup class team may be going a bit too far...
Now, if we suppose that the 2050 goal is not achievable, how does that affect its quality of being worthwhile? Proposing non-achievable goals (or making non-achievable preditions) can be risky. Take the predictions of the Artificial Intelligence community in the early 1960's. Claims were made that within 20 years computers would be as smart, or smarter than, people. Researchers felt they were very close to creating a "learning machine" that could learn much like the human brain. Unfortunately, these claims were way too optimistic and the result was a backlash against AI. AI fell out of favor by the 1980's and only started to rebound in the mid 1990's. In general people don't like to be disappointed, especially the people who are giving out grant money (NSF, DARPA, etc.) Also, the public is easily disappointed by a lack of success, mostly because they have no perception of the difficulty of the problems in any scientific field. And making "wild" predictions is not going to inspire confidence in any field of study.
As for the RoboCup 2050 goal, it has been suggested that it is intended mostly as a motivator for research, not as a seriously attainable goal. Personally, I think achievable goals are better for motivation purposes--if a goal is too "far out" is just not taken seriously. For example, if I proposed a goal of faster-than-light travel by the year 2050 I would be laughed at. Therefore, in my opinion, it really does matter if a goal is within the realm of possibility. The question remains: is the RoboCup 2050 goal achievable?
In order to decide whether any goal is attainable, the key components can be analyzed to see where they fit into these categories:
A) Existing technology,
B) Incremental technology,
C) Order-of-magnitude technology, and
D) Breakthrough technology.
Item A is self explanatory. Item B is an extension of existing technology, usually scaling up or down in size/speed. Item C is a fundamental change in the implementation of a technology that vastly increases functionality but uses the same concepts. Item D is completely new technology.
Examples:
a) Integrated circuit
b) Transistor -> integrated circuit - the IC is essentially just many transistors on a single substrate--this is just a scaling operation
c) Vacuum tube -> transistor - a transistor uses the same concept as the vacuum tube (using a small voltage/current to control a large voltage/current), but the transitor is much smaller, is much more reliable, is much more efficient, and uses a very different implementation
d) Vacuum tube - before the vacuum tube electronic amplification was not possible (the closest thing to an electronic switch was the mechanical relay which was orders of magnitude slower and was not capable of amplification)
Using these categories, it is interesting to look back at JFK's moon landing goal and do the analysis. A sucessful moon landing required these systems:
1) Propulsion (rocket capable of launching a sufficient payload)
2) Control (computer/guidance package capable of knowing the location of the spacecraft controlling rockets at the precise time)
3) Orbital mechanics (math needed to plan a course to the moon and back)
4) Life support (keeping the men alive in space)
5) Miscellaneous
Now the question becomes what technology level was required for each item:
1) Propulsion - B - Germany first flew a ballistic rocket in 1944, the question here was developing a large enough rocket to get a big payload into orbit
2) Control - B - inertial guidance was used by the Germans, computers existed but had to be made small enough and reliable enough for space travel, also the transistor existed
3) Orbital mechanics - B - the math has been available for a long time, computing it was not easy with a slide rule but computers were available which could do the calculations (interesting to note that most, if not all, of the design for the rocket hardware was done by engineers using slide rules!)
4) Life support - B - jet pilots had been using pressurized suits for a long time, modifications were necessary for exposed space travel but the concept was the same
5) Miscellaneous - C - there were some new materials developed for the space program such as the heat shield material for nose cone reentry, various plastics, possibly some new alloys, and Velcro but I don't think I'm willing to call these breakthrough technologies
Using this analysis, the moon landing goal is perhaps not so amazing from a technological perspective. This is not to say that any of the items were easy. An incredible amount of time and effort was needed to design, build, and test the fantastically complex machines required to pull off a successful moon landing, given the less-than-ten-year timeline. I believe the people who were skeptical about the goal didn't believe everything could be done in the time allowed--that just goes to show what several billion dollars can accomplish.
Getting back to RoboCup, what does the analysis look like? In my opinion, a successful World Cup-beating team of robots needs these systems:
1) Actuators (giving it the ability to move)
2) Sensors (vision, tactile, inter-robot communication, hearing?)
3) Low-level motion control (balance, timing, agility)
4) Low-level sensor processing (making sense of noisy sensors)
5) High-level control (goal-based behavior, strategy, learning)
6) Power (it's not eating Wheaties for breakfast)
Now, what technology level is required for each of these systems:
1) Actuators - B/C - current actuators are too slow for the power they are delivering (high gearing) and they are not robust enough to take the punishment of running for 90 minutes; new gel-type actuators which function much like muscles may be the answer but they are only in the prototype stage now; humanoid size constraint makes this one hard
2) Sensors - B - we have good sensors now (not as good as the human eye but possibly good enough), by 2050 we'll have much better ones, full body tactile sensing is being worked on--by 2050 we'll have it
3) Low-level motion control - B - there are good examples of excellent motion control in the robotics community but many of them are low degree-of-freedom applications--dynamic balance & agility have yet to be demonstrated on a humanoid robot, certainly not at anything approaching speeds needed for World Cup soccer competition
4) Low-level sensor processing - B - we're getting better at vision processing but we still aren't near human standards, perhaps throwing 2050-level computation at it will help
5) High-level control - C/D - learning is the real problem--we just don't know how to do it, we can train neural nets to solve some problems but nothing approaching the complexity of soccer (yes NavLab can drive on the road but it's mostly doing line following); I really believe that dynamic, online learning (figuring out what is working/not working and adjusting strategy accordingly) is necessary to play at the World Cup soccer level (unless you've got a robot that is substantially faster & more agile than a human, in which case strategy is not as important)
6) Power - C/D - current power sources are not sufficient (the Honda Humanoid can operate for 15 minutes max and that's at slow walk), I don't think simple extensions of current batteries are good enough--we need some order-of-magnitude changes or some new technology here
So, looking at my analysis I have to conclude that the RoboCup 2050 goal is more of a technological challenge than the moon landing! Of course, we have five times as long to do it so that helps. Maybe some great new technology will come along and simplify things, but it's tough to bet on new technology happening (new technology will happen in the next 50 years but it may be useless in this application). Two additional problems which make the RoboCup 2050 goal even tougher to achieve are money and organization. NASA had the money and had lots of private sector scientists/engineers who were used to working together (and *had* to or they would lose their jobs). RoboCup is not going to get anything like the billions of dollars that NASA had to work with. Furthermore, RoboCup has to depend on lots of academic scientists/engineers working together. This is unlikely from a social perspective (scientists don't always see eye to eye, they are competitors for grant money & publications, and no one is going to make them work together) and from an organizational perspective (scientists are not organized to a common purpose--they don't pool their resources, and they build robots systems which have little in common with other researcher's systems which makes sharing results difficult). Given this context, I think it is unlikely that the RoboCup 2050 goal will be met. Of course, I could be completely wrong and the 2048 World Cup champion team may find itself losing to a robot team. I guess we'll just have to wait and see...