Robot soccer dogs

[Music] Peggy Fidelman
One thing that’s really
great about robotics is that you get to see what you did. I guess maybe everybody wants to have a little army
of slaves doing everything they tell them to do, right? I mean, you tell the robot to walk in a certain way,
and when that works, you get to see it do that. And that’s somehow much more affirming than, you know,
watching your program give the right line of text
as output or something like that. Nate Kohl
Well, the overall goal of
the project on the highest level is to program a group of robots in 50 years, or by 2050
to beat the champion human World Cup team. Our project right here is just to get a
bunch of robotic dogs to play soccer, and to do interesting research while we’re at it. Peter Stone
I’ve been involved in
the international robot soccer initiative, known as RoboCup, since the beginning, back in about 1996. This has now turned into a very large international
initiative involving, I’d say, over
3,000 or 4,000 researchers worldwide. Our project started last January, and we had a first version of everything
working by about the middle of April. But, this was probably the quickest it’s ever been done. This was started as a class with 19 graduate students
involved and they all worked very hard to get a
complete solution within those few months. Nate Kohl
The software that
controls whether it’s playing offense or defense is all stuff that we’ve coded up, various different criteria and
decision-making kinds of things. Peggy Fidelman
The first thing we
try to do is we try to get it to score on a goal. That’s actually very hard. First you try to make it walk and
have some sort of kick kind of thing, to be able to find the ball. Then, you start working on having
the multiple robots cooperate. So, instead of just having four
individuals go right to the ball and
run into each other on the way there, you have them, sort of, decide among themselves
who’s going to go toward the ball, who’s going
to get near the ball, but not go right to it. Peter Stone
Every year we take our team
to an international competition and work against other people around the world who have been working in exactly the same domain
in a head-to-head format so that we can compare our
results and compare notes and learn from each other. [Audience cheering] Nate Kohl
The overall sense of the game is a large crowd of people clustered around a tiny field and all these people are screaming like mad at a
bunch of little robotic dogs that are all piled up
trying to get a ball into a goal. The games themselves are usually pretty exciting, too,
because you never know what a team’s going to come up with. Like last year, the German team came up with
a pretty cool kick where they actually grabbed
the ball and threw it over their head backwards. [Crowd noises] Peggy Fidelman
Once the dogs start playing we don’t have any say at all about what they do. They’re just out there and sometimes we’re just
watching them and we’re like, “No, don’t do that!” There’s nothing we can do. Peter Stone
One of the goals of our
research is to allow the robots to adapt or learn to change their behavior as they
go during the course of a game. One of the big successes we have right now is
the robots have learned to walk much more quickly
than we were able to hand-code them. Nate Kohl
What we did is we had them…
We had a couple of robots on the field and they just ran back and forth across the field over and over and over and they
timed themselves as they went. So they were like, “Ok, I crossed the field
in two seconds. That was really good.
I crossed the field in 30 minutes. That was bad.” Sometimes you’d come in here at night and
the robots would be training. They’d just
be pacing back and forth across the field. They would learn, basically, what walks
were fast and what walks weren’t fast. We got really good results.
It was 291 millimeters per second. Which, at the time, was the fastest walk known for an AIBO. Peter Stone
The goal of having robots beat the
best World Cup soccer team on a real soccer field is not in and of itself necessarily a benefit to society, unless you take the entertainment perspective. But, in order to achieve this goal, there’s many basic
spin-off technologies that will be necessary. There’ll be things that apply to many real-world tasks, perhaps someday eliminating the need for
traffic lights because the cars can negotiate
among themselves as they approach intersections. There’s a direct application of some of the
work in the disaster rescue scenario where the idea is that robots are helping people rescue
victims of natural disasters such as earthquakes. The basic issues that we’re examining in this robot
soccer domain are going to speak to that. Nate Kohl
It’s an interesting thing, you know? It’s kind of a new take on lots
of the classic problems of AI (artificial intelligence). It’s a new way to look at them. Peggy Fidelman
It’s very addictive work.
That’s the thing about robotics. People either don’t like it at all or
just get so into it that they’re really
wanting to finish all their other work so they
can get into the lab and work on the robots some more. And that’s sort of what happened to me.

Leave a Reply

Your email address will not be published. Required fields are marked *