How can robots that do not know each other cooperate? Peter Stone from the University of Texas and Israeli researchers are working on the answer
by Eitan Crane
It's a cool midnight hour in Jerusalem, the striker in yellow and black grabs the ball, breaks into the wing, lifts it into the box and... a goal! No, Beitar Jerusalem did not score a goal and the "Teddy" stadium in the capital did not burst into roars, the soccer players in front of my eyes are small circles running on the computer screen of Professor Peter Stone from the Department of Computer Science at the University of Texas at Austin.
Professor Stone is in Israel on a Fulbright Program scholarship, which is given to outstanding researchers in their field from Israel and the US, and is managed by the United States-Israel Education Foundation. Each player on the field is controlled by a separate and independent software that decides the moves of the "player", explains Stone. But Stone aspires to more than that. He is interested in bringing the artificial intelligence (AI) involved in a football game, or any other human activity, to the turf of real reality. Indeed, while the visualization on the screen looks quite realistic, the next videos Stone showed me, showing robots playing football, demonstrated just how difficult the task still is. Even with the programs making correct decisions, the robots, in their variety of forms, sometimes have difficulty performing the simplest actions. The puppy-like goalkeeper sometimes scores his goal, and the two-legged robots tend to trip and fall. Although things look clumsy, the progress in this field of research is great, Stone says and demonstrates his words with videos showing that just a few years ago robots looked much clumsier.
Peter Stone's field of research, which combines artificial intelligence and robotics, faces fascinating challenges. There are researchers who deal with artificial intelligence using only software, but they ignore the fact that the software installed in the robot must deal with reality. She must also take into account the robot's physical limitations, and especially the sensory input based on which it makes decisions. But Stone goes even further and examines how to build multi-agent systems (MAS) in which different robots interact with each other, learn new situations together and function effectively as a team.
Soccer, robotics and Israel are integrated topics in Stone's life. He caught the football bug in his youth when he was in Israel with his family. When he returned to the US he continued to play soccer as a hobby. Then, in 1994, he met researchers from British Columbia at a scientific conference who were demonstrating two robots playing soccer against each other. Although the demonstration was impressive, Stone thought that "real football is a team game" and decided to devote his doctoral thesis to building a team game through the combination of two areas of computer science: machine learning and multi-player systems. This is also how Stone got to know the RoboCup league where robot teams compete for the soccer world cup with the goal of beating the team that will win the world championship in 2050.
Why football specifically? First because it is truly an international game that combines team play and individual ability. But also in terms of computer science, football has an advantage. Unlike other ball games, such as basketball and volleyball, soccer is an almost two-dimensional game. And will they really be able to beat the world champion? The main problem is mainly in the engineering of the robots, here we are really far from competing with humans. But, "70 years after the Wright brothers' first flight we reached the moon, 50 years after the discovery of DNA the sequence of the human genome was determined, so it seems that 60 years after the founding of the RoboCup league it will be possible, perhaps, to reach this," hopes Stone.
Soccer is just one example with defined rules for robots working in a team. But even in the real world, Stone expects to see robots soon: driverless cars, home robots that will do our daily chores for us, and robotic rescue and rescue systems that can operate in disaster areas where it is too dangerous to let humans in.
In such situations, robots that have not met before will meet each other, yet will have to work as a team and not interfere with each other. It is this new and fascinating field that brought Stone to Israel. "This is a new field in artificial intelligence, which combines social choice theory with game theory. Together with Professor Gal Kaminka from the Computer Science Department at Bar-Ilan University, who is one of the leaders in Israel in the field of artificial intelligence, Stone is designing a robot that is a "good team member" capable of dealing with robots that are better than him or inferior to him. Like a person who meets another person who does not speak his language, yet they have to work together, the robot also cannot directly influence the foreign robot, because it cannot communicate with it and it does not know its software. The robot, like a person in his situation, can only control his own actions. He will therefore have to decide which actions will make the alien robot cooperate. In the meantime, Stone and Kaminka are examining "one-on-one" systems. Their joint research, says Kaminka, "will allow robots that do not know how to communicate with each other, and that have not met before, to cooperate optimally."
His other Israeli partners work with Stone on the social decisions the robots must make. "In real situations the robots must respond to everything that happens and not just be satisfied with what makes sense to do," says Stone. And in Israel there are experts on these issues, such as Professor Jeffrey Rosenshein, head of the multi-participant systems research group at the Hebrew University of Jerusalem, Professor Sharit Kraus and Dr. David Sarna, both from the Computer Science Department at Bar Ilan University. Much work has been done in Israel on theoretical issues such as social decisions, voting in many participatory systems, software capable of investing in the stock market. "I brought my experience in machine learning," says Stone. He hopes to establish a research channel here that will focus on the question of what characterizes a "good team member" and will be based on the joint expertise of the researchers. "This is the academic reason for my stay here, but I also wanted my children to get to know Israel better, as I got to know it in my youth," says Stone in Hebrew.
An example of a robot that is a "good team member" is the robotic car, another subject that preoccupies Stone. In a future world where the cars will drive themselves, the robotic car will have to operate against other robots it has not encountered before, i.e. against other robotic cars of a different make. Stone shows me a simulation in which hundreds of cars cross an unregulated intersection of two multi-lane highways without colliding. Each car informs the central computer that it is approaching an intersection and the computer assigns it a reserved lane. The few cars that did not receive such a lane, wait for their turn in a special lane.
But even here Stone is not satisfied with software simulations: together with his colleagues in the "Texas Robot Technology" team, he participated with the Isuzu autonomous robotic vehicle "Marvin" in the "Urban Challenge" competition of the American Agency for Advanced Scientific Research (DARPA) in 2007. In the videos that Stone showed me, the car is seen when it reaches the intersection, giving way to the cars that came before it, but taking the right of way for itself and turning perfectly to the left, not before making sure that the driver in front of him is not going to explode. "Do you trust a car like that?" I ask, and Professor Stone reminds me that in a normal commercial flight an autopilot is in control most of the time.
"What is intelligence anyway?" I asked Stone. "There is a phenomological definition," he replied. "If it does things that are considered intelligent then it has intelligence. It was once thought that a computer that could beat a human at chess would be intelligent. Today there is such a computer and there is no consensus that it is truly intelligent. So the definition is elusive.” It seems that Stone is moving in this direction step by step - to create computers that are able to do new things that were not able to do until now, while integrating sensing means.
In conclusion, I ask Stone's opinion if such a robotic world might not take over our lives like the computer HAL took over the spaceship in Arthur C. Clark's book "2001: A Space Odyssey". "The ethical question is indeed beginning to arise, and there is no doubt that in the future the legal system will have to enter into this," says Stone. Asimov has given the subject deep thought in the past, but it seems we are getting closer to the day when the subject leaves the pages of science fiction.
Dr. Eitan Crane is the Scientific-Operational Editor of Scientific American Israel and heads the Culture-Science Program at Hamada - the center for scientific education in Tel Aviv.
For the interview that Stone conducted with the science website in January - By 2050 robots will beat the world champion in soccer
