By 2050, robots will be able to compete in a soccer match against the world champion

Says Peter Stone, professor of computer science, who specializes in artificial intelligence, at the University of Texas at Austin, who is spending a sabbatical year in Israel and is researching, funded by a Fulbright scholarship, together with four Israeli researchers, various aspects of the field of robotics * Stone teaches the robots to understand the world and cooperate with other robots

How close are we to Star Wars robots? To a certain extent, for over thirty years after the release of the first film (well, the fourth) in the series, until films like "I am a robot" that came out a few years ago, we see robots in the cinema whose behavior, even if not always uplifting, is human.

These films, and the television programs dealing with robotics are the worst enemies of researchers in the field because people believe that robots in reality are already analyzing the world, drawing conclusions, moving as they should, cooperating, while in most fields the situation is far from that. To get to robots like R2D2 or C3PO you have to go a long way in everything that makes the robot what it is. They are very far from understanding natural language or drawing conclusions and their vision is limited due to the ability to decipher, they are barely able to manage to walk on two legs, and it is very difficult for them to cooperate. This is how Prof. Peter Stone from the University of Texas at Austin, who is doing a sabbatical year in Israel, which he spends in research together with Gal Kaminka, Sharit Kraus and Dodi Sarni in Bar Ilan and Jeff Rosenshein from the Hebrew University.

The areas that Stone researches relate to the issues of artificial intelligence and cooperation between robots. "I am a software person, but I also need to know the body into which I insert the software, so I program the actions of specific robots."

Robots know very well how to calculate, get an equation and solve it, like any computer, but it is impossible to talk to them in the natural language - Hebrew or English. You cannot let the computer read a book and then answer questions about it, as children in the fifth grade do.

Stone's research interests range from artificial vision, through playing soccer with robots, to an autonomous car. Each of these areas is a whole world. What they all have in common is that in every field on their way, the robots have to face challenges and the more they advance in all three stages - perception and understanding, movement and cooperation with other robots, the better.

The challenges are many. For example, the field of artificial vision. Stone explains: "Robots still can't see the world as well as we can. If a person can look at a picture and say here is a computer, here is a chair and there is a person sitting (and even a specific person), because we look and see and immediately understand everything. A computer cannot sense the world in the same way. The computer receives pictures taken with a digital camera. These images are a file containing groups of three numbers and the intensity of the red, blue and green color of the dot. An image for the computer is a long list of numbers. A challenge for computer science or artificial intelligence is to turn the line of numbers into an understanding of the picture - where are the orange things, where is the chair. It's not easy, but there is progress."

"Another challenge we are far from is making robots capable of understanding and communicating naturally with humans. I can look at you and know if you are sad or happy and treat you accordingly. Computers and robots cannot do this. We are still far from robots like the robots of Star Wars. An amazing development has been made, but it is still science fiction.”

Robot Soccer World Championship

An important part of the ability of a robot like C3PO is learning not to repeat mistakes. The problem is that every situation is different. You should never repeat the same decision, so it is difficult to know exactly what lessons should be learned from the past. This is true for humans as well as robots. An important part of machine learning is determining how to represent the current state and relate it to past states. Our challenge is to know how to react in similar but not identical situations and unlike the common stereotype about robots that they are always programmed to do the same thing, the robot needs to be taught generalization - to know how to react in similar but not identical situations.

Besides the ability to make decisions in situations that humans are faced with for example, we want to have robots that communicate with each other and with humans. For example, I am building a car that will drive itself (and there is one in America that I am developing). Many cars will actually travel on the road and they will need to communicate with each other. The ability to see and the ability to cooperate with other robots also comes in the World Championship for Robots in Soccer. With most of the Israeli researchers, I collaborate in studies in the field of Multy Agent Systems - a computer program that can communicate with other programs of robots that are teammates or, in the case of football, of the opposing players. A test area for me is the robot playing soccer, which can communicate both with the robots that are its teammates and with members of the opposing team. This is something that can demonstrate how a multi-agent system works.

Together with Gal Kaminka, I participated in the World Robot Soccer Championship - RoboCup, I myself have been participating in it with my team from Texas for ten years and we brought many medals, in the leagues divided according to the types of robots. We bring our robots and compete against teams from other places. It is good to develop machine learning, collaboration and also the professional sides of robotics. In one of the competitions there was also a championship for a game between humans and a group of robots moving on wheels one meter high built by developers from Germany. I scored two goals for the human team and then they asked me to play for the robots and I scored goals for them as well. Obviously the robots can't do what we can do. They still can't see well and decipher what they see, they can't move on grass and can't predict what the opposing team is doing.

"As mentioned, the goal is to reach the ability of robots to beat humans in 2050, but even if the robots only beat teams from the third league in Israel and not the world champion, this will still be a sign that we have overcome many challenges." states Prof. Stone.

What is holding back the autonomous car is the price of the sensors

Another pride of Stone is designing a car that takes into account road conditions, other cars and traffic signs to drive on its own, thereby reducing the risk of human involvement in driving. "Most of the flight time is carried out by an automatic pilot, and it is even possible to land in this way. But people are not afraid of airplanes without pilots, but they are afraid of driving a car without a driver.

Last year we built, based on a 1999 Isuzu Viacross car. In the example in the video (see link) the car reaches a stop line, and there are four cars waiting on several sides of the intersection. According to the law, the order of cars leaving the intersection is according to the order of entry. The car needs to sense the world and the rules of the road have been designed for it. She needs to recognize when it's her turn to go to the intersection, which cars she should consider, and which ones she shouldn't (for example, those who are signaling towards her so as not to disturb our car). When she comes to the conclusion that she can go out, she does go out to the intersection and does so safely. We designed her so that she can also know when it is safe enough to turn left through the traffic, and even bypass a block by reversing, turning in the nearest block and bypassing it, and arriving at her destination. The car also has all three components, sensing and understanding, cooperation with the other cars, and making the movements according to the plan."

What is the reason why we still do not have autonomous cars today?

There are not many reasons for this. The technology is mature, if once you needed a supercomputer like in the cars demonstrated in the nineties, today two ordinary computers that you buy in any store are enough. The ability to control the machine and carry out the commands (drive, stop, turn, etc.) is already good enough. What still needs to be improved is the sensing. In order for us to reach our performance, the sensors are more expensive than the car. There is a need for the production capacity of cheap and reliable sensors, because here the reliability must be XNUMX percent.

By the way, in the same matter, Stone shows us a demonstration video that shows that if all the cars were autonomous, there would be no need for traffic lights on the road. In this video you can see complete chaos - cars are constantly entering the intersection from all directions and some are turning and everything looks like a big mess, but it turns out that no accident will happen because the computer of the intersection acts as a router, and approves or disapproves each car to enter the intersection when the cars transmit to it where they want to go.

As mentioned, Prof. Stone is in Israel on a Fulbright scholarship, which is given to outstanding researchers in their field from Israel and the US, who come for ten months to research together with their colleagues in a leading department and it turns out that Israel has quite a few leading departments in many fields.

The United States-Israel Educational Foundation, established by the US and Israeli governments in 1956, is responsible for managing Israeli participation in the US government's Fulbright program. Fulbright is one of the most famous and most expensive programs in the world for student and lecturer exchanges.

The main goal of the Fulbright Program, which was initiated by Senator William Fulbright in 1946, is to strengthen the foundations of peace by strengthening the understanding between the American people and the peoples of partner countries around the world. The main means of achieving this goal is the exchange of students and lecturers at the highest level of academic excellence.

In the years since the founding of the US-Israel Education Foundation, over 1,100 Americans and over 1,500 Israelis have participated in a variety of Fulbright program lecturer and student exchanges.

"I came to Israel for a period of one year, which I am already in the middle of. I managed to develop cooperation with my friends in Israel and when I return to America we will continue to keep in touch and move forward."

video clips

• The movie in this link shows the robots walking fast and playing with a ball
• A short explanation of the robot - video
• Several links from the competitions appear
• The smart vehicle in action