Robots That Jump

Robot Bodies Needed Before Robot Minds

Robots That Jump – Historical, June 8-12, 2003

Thursday, June 12, 2003
Why fun, jumping robots are better than the “dull, dirty and dangerous” mantra
Isn’t it more important to develop ultra-reliable machines that can carry out tasks with the realiability of and industrial robot? Shouldn’t all robot research be devoted exclusively to things that matter – e.g., dull, dangerous and dirty jobs? Isn’t it more important to fit robots to a specific useful function than trying to replicate immensely complex general tasks?

Wrong on all counts. There are many reasons for the “long winter” of artificial intelligence and robotics (noted by robot pioneer Hans Moravec), but the above attitudes are at least part of the reason robots haven’t appeared yet. Here is a consideration of each argument, and a response, leading to the conclusion that robots really should jump around for our entertainment:

The “dull” argument says that we should create ultra-reliable machines to automate boring tasks. Since the task is dull, a general-purpose robot isn’t needed. I saw this position taken most recently on a Slashdot posting after Robodex 2003 in Japan. The poster claimed that it was a mistake to create complex robots with general-purpose bodies. Instead, they pointed to the industrial robot world. Industrial robots started simple and have actually become simpler in many ways. As industrial robots developed, it was found that simple, inflexible controllers were superior to complex systems trying to integrate senses to control movement. Instead of making robot arms with the flexibility of a human arm, the industry restricted degrees of freedom and focused on making the limited motions as exact as possible. Since flexibility is a problem, the arms must be made out of massive metal components to impart absolute rigidity.

The writer suggested that this evolutionary path for industrial robots proved that general-purpose robots were a pipe dream.

The problem with this idea is that industrial robots have changed little since the 1980s – and their adoption has equally slow. In the early 1980s Time magazine ran a cover story on the “steel collar workforce” predicting widespread robot adoption in a few years. Over 20 years later, industrial robots are doing nearly the same things they were doing then, which is fixed operations in an ultra-controlled assembly line environment.

Why? it is the easy way. I don’t mean it isn’t very difficult to create a reliable commercial machine like an industrial robot — but making a huge, inflexible, insensitive robot arm is easier than making a small flexible, sensitive one that does the same thing. Programming a series of simple, inflexible motions for the arm is easier than creating an arm capable of recalculating each motion using current sensory and motor information to accomplish the same thing. The end result is a device incapable of adapting to more complex environments.

A similar problem exists with confining robots to dangerous jobs. If a robot is involved in bomb removal who is going to trust its actions to itself? The overriding concern for safety requires that it be a big electric puppet under absolute human control. All efforts are given to getting the machine under ever-tighter human operation and none to making it responsive to sensation or planning its actions. Any sensory data is fed back to the human operator. The machine itself remains a dumb puppet with no ability to respond to its environment.

Once again, this is the “easy” way out. It’s hard to make tele-operated systems, but still far easier than making these systems capable of handling dull, dirty or dangerous environments on their own. A tele-operated robot may be given some autonomy (e.g., the NASA MER robots), but they can’t be allowed to really run on their own unless their “perfect.” There is a huge gap between the near-zero autonomy of a tele-operated system and the super-competence required to remove the human from absolute control. Companies creating tele-operated robots for dangerous jobs will be unable to make this jump, and have little reason to do so.

Finally, some suggest that creating special-purpose robots is a better goal than general-purpose robots. Instead of creating a robot that drives a car, why not make an “intelligent” car. The problem here seems to be an implicit assumption that by making a special-purpose device, the complexity of the environment it needs to understand is simplified. A vacuum cleaner only has to understand the “vacuum” environment, for example

However, practically any “natual” environment (meaning it hasn’t been structured explicitly for robots like an automated assembly line) has enormous complexity. To really vacuum well, a robo-vac should be able to determine if the carpet is dirty, and if it is successfully cleaning it as well. This is clearly a very advanced function. Failing this, a human must expend effort making these determinations or the robot must vacuum far more than it is really necessary to “be sure.”

A robo-vac also needs to recognize people, furniture and pets to work without supervision. Making the robo-vac small, round, soft is an “easy” solution that leads to a dead-end performance. Problems will still happen. In contrast, wouldn’t it be great if a general-purpose device that “knows” what furniture and pets are?

Why entertainment: Unlike these other domains, entertainment has some unique features that demand real robots. In these environments, perfect accomplishments aren’t necessary – the robot can be clumsy or playful and still be entertaining. At the same time, real, lifelike behavior will be rewarded by its public. So the entertainment robot developer has a smooth path to more advanced, general purpose behavior. Each robot they create along the curve in increasing ability can be entertaining. Early versions of a general-purpose system will be entertaining, whereas a clumsy bomb-removal robot is useless. A robo-dog that can just barely play soccer a la RoboCup is entertaining, whereas a robo-vac that doesn’t work well is simply junk.

At the same time, the very frivolity of entertainment provides a push to increased robot performance. A cutesy, clumsy entertainment robot is fun, but before long the public tires and wants more fun. This isn’t necessity, just fashion, so it isn’t as serious as a lousy robo-vac, an inefficient welding arm, or a pet-scaring robo-vacuum. At the same time, the pressure is there.

All in all, entertainment robots provide a smooth path to increased robot performance, without any abrupt jumps in ability being necessary. And the end result – something that can dance and clown around – will have the general-purpose intelligence needed to do the dull, dirty and dangerous jobs and really integrate robots with society.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: