UNSW competed in the Sony 4-Legged League from 1999 to 2006 with amazing results, becoming World Champions three times, 2nd three times and 3rd once. They have also sustained these results being World Champions in 2014 and 2015 in the International RoboCup League
Mike Lawther remembers how UNSW became involved in the competition:
“Back in late 1998, Will Uther came out to UNSW to give a talk on the work he had done with the prototype dogs for RoboCup-98. The dogs were an experimental event in 98 – but in 99 they were to become an official league, and Sony were looking for expressions of interest.
We (as in John and me) told Claude that we thought this would make an intellectually stimulating (read cool and funky) thesis project. So we needed to write a proposal to Sony about who would be involved, and what we proposed to do. So we put together a proposal saying all the great machine learning stuff we planned to do, and that there would be a high powered team of brains behind it, including Claude, Graham Mann, Waleed Kadous, Malcolm Ryan, Jamie Westendorp, Phil Preston and John and me. Sony accepted our proposal in January (yay!) and it was all good to go. John and I got into reading everything dog related and RoboCup related that we could lay our hands on. And we planned the victory dances the dogs would do when they scored goals.”
Indeed it was a great feat to be selected – 24 teams from universities around the world applied, but only nine were selected.
UNSW was the only team from the Southern Hemisphere.
“It wasn’t until April or so that we got our first dog (quickly christened StupidDog, because of all the trouble we had making him do anything!). StupidDog was really cool looking though – just the insides, no plastic skin, and things held in place with gaffer tape. It was probably May by the time we had the little bugger walking around. Sony documentation seemed slow in coming (we reckon that we were getting it as soon as it was being written). Recall that the competition was at the end of July. All of the great proposal went out of the window as we tried to make things work. Back in those days, we didn’t have a whole lab to work in – we got ourselves a small wedge shaped office, and we could set up a quarter of the field in it. We worked like this for a while, until the break at the end of session 1 when we moved into the robot lab. We had about a month to go, and this was the first time we had set up the full field (this lab had a big orange door, which tended to confuse the heck out of the poor little things, as they kept seeing what looked like a huge ball off in the distance).”
“By the time we had left for Stockholm, we had a striker that could eventually get a ball into the goal, and a goalie who would have probably been more effective if he was switched off. In the few days of competition, we managed to rewrite the goalie completely, tweak the striker, and retune our vision system. Luckily we had Phil there as our team manager (or mum) to make sure we ate and slept occasionally.”
Mike and John excelled beyond solving these teething problems (although some like the orange door still exist!) to come second at Stockholm. From their groundwork UNSW has been able to become a major player in this competition and continue to have great success. The following year Bernhard Hengst invented the P-Walk, which gives the dogs a low center of gravity and is over 300% faster than Sony’s walk! All of the strong teams in the competition emulate this walk now.
With the competition being “Open Source” (the source code is published at the end of the year so that AI improves all around the world), UNSW has to work hard at not only gain an edge, but to also out-do themselves each year. The core team of UNSW RoboCup continues to use fourth-year undergrad thesis students, all of whom have to learn about Robocup’s technology from scratch before starting a new campaign.