Translated by Harold B. Gill

The smart chap from Bonn

By Thomas J. Schult

In the University of the American city of Seattle, a cleaning column emerged last summer, to clean a few offices whose floor was full of balls of paper, Cola bottles and other bits of trash from the research. It was an astonishing performance.

Out of 18 cleaning machines that were announced about half showed. Those that came had problems in the smallest tasks. Thus many janitors didn't find the door or rammed into filing cabinets. Others suddenly stopped and stood apathetically and even stayed that way. A few had no arms and had to be helped in collecting each piece of trash; some brought the paper balls to the wastebasket only. One never took the direct path but always hugged the wall. Others constantly grabbed at the shadows of a coke bottle and mistook a tablecloth for the trash barrel.

Four of the cleaning professionals passed, only in clearing away garbage that they themselves had brought. Even in this, though, they needed help! They required that the trash be laid visibly onto their base. Only then would they take it along.

Even so, hundreds of people watched the spectacle with excitement. It was a type of world championship in which the best in the field were entered. The incapability that was hardly deniable exhibited by the participants was due to one detail beyond all others. The entrants were not human. They were robots; autonomous robots (so-called because they are capable of autonomous movement) to be precise.

These creations are to have long ago taken the tedious tasks of everyday life off of our hands. That will take a little more time, however, as was highlighted by the robot competition in Seattle. The drivers in the artificial office space created there made it clear that machines cannot leave the work place cleaner than most humans.

On of the few bright points in the darkness of artificial agility was a robot by the name of Rhino. It achieved second place. Anyone who wants to visit Rhino in its home city must come to the Institute for Information Sciences (Institut fuer Informatik) at the University of Bonn. Rhino came at the end of 1993 as a brainless piece of hardware to this place just as the Rhein was threatening to drown the city - hence the name. Since then the machine has learned a great deal. One can cross its path in the Institute hallways.

As a chest high black barrel on wheels, Rhino may not cut a dashing figure but it is almost possible to describes its actions in human terms, so curiously does it roll into all the corners and niches. After it has looked into the ends of the human sufficiently, it accelerates and rushes at full speed up to the visitor - one of those moments since it remembers its human goal, that software is never free of errors.

But Rhino does what it ought to. A meter in front of the guest it puts on the brakes and comes forward more slowly looking with its camera up and down and to the side. A small turn and it rolls by on the right as the author of this goal is classified as "column, free standing" and is forever recorded as a thick black dot in Rhinos internal navigation map.

A door stands open. Rhino rolls through it resolutely, rolls around a conference table and the chair on which one of his creators sits and moves past a desk to the window. Then it is the steering is over. Sebastian Thrun, who gave the cheery barrel this type of life gives it a tap on the shoulder. There near the two cameras Rhino has an emergency switch that leaves him still. If it were to be mad autonomous and endangered people or furniture a hearty clap on the shoulder will suffice.

At an age when others brood over their degree work, the 27 year old Thrun already has two positions one in Bonn and the other at the American Robot-Mecca of Pittsburgh, at the Carnegie Mellon University. For this, he opts fur something higher in conversation. Thrun had rejected the invitation of a New York head-hunter for a position in financial market analysis. There, experts in the application of self-learning neural nets are in high demand. The position would have earned Thrun 250,000 Marks a year but it would have torn him from the circle of his beloved autonomous machines. And what robot fan could withstand a possible offer from Massachusetts Institute of Technology or from Stanford University?

Contrary to all stereotypes of Techno-freaks, Sebastian Thrun is rather a research version of Thomas Gottschalk: smart, eloquent, witty, only without long hair. Thrun needs the US scene for inspiration, at least. Here in Germany everyone jumps over the pond, he says, and for good reason, because the working conditions of elite American universities are not to be found in Europe.

To clear up misunderstanding: In Bonn there are no backwoodsmen working. The professors Armin B. Cremers and Joachim Buhmann have gathered a group of scientist around them for some time who cover every possible field of artificial intelligence. The Robotics can first at the end of 1993 when Sebastian Thrun appeared in Bonn after a half a year in Pittsburgh and the Institute was able to spend 80,000 Marks for an autonomous robot. Thrun brought Know-how along and the enthusiasm for the robot competition mentioned earlier in which he had already taken part with his American research group.

Competitions for machines have since become numerous, particularly in the USA. The goal set for these are varied. Sometimes it involves a high tech cock fight in which the robots strafe along side each other until only one remains; a fest for the public but no challenge for the serious inventor. Since the machines in the competition will find no sponsor due to sinking equipment funding for the conduct of the contest.

Among the serious contests, the robot competition of the American Association for Artificial Intelligence is the most renowned. It takes place yearly. The invitational meet in Seattle was the third in general and the first that also saw foreign entrants. The next will be next August 20 of this year in Montreal: the Bonn team will be there again.

The sheet metal bolides seldom come from the workshops of the participants themselves. The two US companies, Real World Interface and Nomadics compete for the small market. Robot researchers build the necessary hardware. They deliver the autonomous barrels only with the most basic steering software. The machines understand program commands like "Turn 20 degrees." Or report that the ultrasound sensor 13 is in front of an obstacle. Everything else, the purchaser has to install: capabilities like recognition of outlines, production of maps, planning routes, learning from mistakes.

Therefore there are many possibilities for Artificial Intelligence research. The annual competition involves the most effective software. That can't be bought anywhere.

The Bonn program runs on two ordinary personal computers that drive Rhino and on a work-station with which it is connected via modem. In the 6 months since the last competition, Thrun and four colleagues are only finished with a part of the planned navigation software. Many current Artificial Intelligence techniques were already contained, neural nets, case based conclusions, machine learning, that brings about placement in the competition.

Only Rhino was chosen by a US research group from Georgia Institute of Technology that sent four identical robots in a row under the names of four blizzards. They were so coordinated in their software that they did not come into an enclosure. Team robotics are called the research branch. The blizzards were those that needed the basin for the trash that also was their own. The mature point departure they made more than even by their speed.

Rhino was one of the few robots that needed no help - no rough pre-generated map of the office, no familiar trash, no garbage sculptures. Rhino recognized everything alone and very quickly. Not one single blizzard would grow up to be him.

This success made the company Real World Interface take notice of the Bonn Designers. They donated a part of Rhinos steering programs to give the robots as standard software from then on. There was not money for this because the research apparatus does not achieve a high number of pieces but Rhino now got an arm as counter-performance.

Competitions are not to remain Rhino's life work. "We want to create a robot that will reside in the university 24 hours a day," Thrun says, "It should interact with humans and take over all sorts of tasks, like carrying out the mail, transmitting messages, taking out the garbage." In doing this it should also be constantly accessible in the world wide computer network Internet via modem. Rhino's creators therefore are seeking partners in Industry and Promotion from public funds. Contributors are naturally fewer for the runner up to the world championship as interested in practical application. There is work for autonomous robots most certainly, overall where humans are not able or do not wish to go but have something to dispatch from the research into a volcano's crater to the irradiated spaces in a nuclear power plant.

The one in Bonn and also in other places the application envisioned is more base, namely the sewer pipes. Passages in the underground sometimes leak noticeably. At the time it is not known where the tubes actually lie, not to mention where the damaged places are. An electronic crawling animal with its own location capability could help there. Perhaps we will soon be able to think with satisfaction that these mechanical friends crawl out on regular inspection tours in our pipes. That they get very dirty in doing so leaves them completely cold in accordance with their program.