Robot Hockey

Project Start: January 2012

Project End: April 2012

Presented at ICRA Humanoid Application Challenge, May 2012 in Minneapolis-St Paul, USA

Won first place

Our submission video for the 2012 ICRA Humanoid Application Challenge

Background

Dr Jacky Baltes, my Master's thesis advisor was a member of the West German Olympic speed skating team at the Calgary and Sarajevo games. For years he had wanted to try making a humanoid robot skate, but up until relatively recently the technology was simply not advanced enough for such a feat.

When Robotis, the company that manufactures the DARwIn-OP humanoid robots used by the Autonomous Agents Laboratory announced that they would be holding a challenge at the 2012 ICRA conference in Minneapolis, we decided that the time was right to tackle this project.

The rules of the challenge were simple; devise a novel problem to be solved with humanoid robots, implement a solution for this problem using a DARwIn-OP robot (the requirement to use a DARwIn-OP was dropped in later years). To enter the competition you teams needed to upload a short video demonstrating their research. Based on the quality of the videos and novelty of the problems finalists would be selected to present their projects at the conference. Winners were selected by a panel of robotics experts and by peer-review from other finalists.

Early Tests

With only a few weeks to create our video, the pressure was on! Jacky and I quickly manufactured some simple skates out of scrap sheet aluminium in the lab, hand-cut with tin-snips. We has no idea if the skates would even glide on the ice, let-alone whether or not the robot would be able to move and balance on skates.

Fortunately every winter the university floods the quadrangle (a small field) just outside the building in which our lab was located to make an outdoor skating rink, so it was a short walk to get to some ice.

With Jennifer, our brand-new DARwIn-OP robot -- named after Winnipeg native and multiple Olympic gold medalist and world champion in women's ice hockey, Jennifer Botterill -- in tow, hand-made skates bolted to her feet, Jacky and I braved the -20C for our first test.

The initial test was simply putting the robot in a static, standing position with the feet parallel and pushing the robot on the ice to see if the skates could glide.

This first test was a tempered success; we made the skates too short and too tall; the robot had very little stability front-back, and would easily fall over face-first onto the ice. However, the basic gliding principle was demonstrated; we could make robot-sized skates that would glide on ice.

After some minor modifications to the skates we tried our pushing experiment a second time, with more success. The shorter, longer skates provided adequate stability in the front-back direction, and could still glide on the ice.

Self-Propulsion

While pushing the robot with our own hands was adequate for a test, clearly the robot was not skating yet. Fortunately, as our silver medal in the 2011 HuroCup Marathon demonstrated, the DARwIn-OP robots can walk on smooth ground. Ice is fairly flat and smooth, so why not use the robot's own parametric walking module for skating?

We modified the robot's walking gait slightly to use lower, longer strides, and set the robot "walking" on its skates. It took some time to get the parameters tuned to the point where it was actually able to make forward progress, but we eventually had something that looked like a three-year-old child carefully shuffling on skates. For the first time Jacky and I thought that we actually had a shot at making this work -- and it had only been a couple of weeks.

One of Jennifer's first tries at skating on outdoor ice

Hockey, Because Canada

When most people think of Canada, the first thing they tend to think of is very often hockey. We had demonstrated that Jennifer the robot could -- by the crudest of definitions -- skate. The next logical step was to put a hockey stick in her hand and get her to play hockey.

I freely admit that the hockey modifications were done largely for the sake of making a better video; there was very little useful research to be done with simply slapping a puck or ball. However, thematically it simply felt right to make the robot play hockey.

Using some basic computer vision algorithms to track a red ball and programmed Jennifer to hit the ball with either a forehand or backhand slap depending on what side of the stick the ball was on. No localization, no aiming, simply reactively hitting the ball when it intersects with her path.

(We tried using a hockey puck instead of a light foam ball, but the robot's shoulder motors were not strong enough and the puck was too heavy for this to be effective. Also, despite being right handed myself, I always found it more comfortable to hold a hockey stick left-handed. Therefore Jennifer was programmed to use her stick in the left hand.)

The Video

After refining the forehand and backhand shots more, and improving the quality of the robot's shuffling skating gait, we were rapidly approaching the deadline for competition entries.

With a Team Canada hockey uniform from Build-a-Bear Workshop -- provided by my mother -- we took Jennifer back to our outdoor skating rink and spent several hours filming her skating and shooting from all angles. The temperature was beween -25C and -30C, putting a toll on our computers, batteries, and bodies. But in the end, as the sun was setting, we packed up with enough footage to put together our application video.

I spent a couple of days editing everything together, all set to the Hockey Night in Canada theme song -- a piece of music that is as iconic in Canada as hockey itself, but is largely unknown in the rest of the world.

Media Firestorm

For reasons that I cannot explain, the video exploded online. "Exploded" is perhaps an over-statement, but it rapidly hit 50k views, the most any video of mine had ever been watched, and got noticed by assorted national and international media outlets. We were interviewed on CBC, CTV, the National Post, the Winnipeg Free Press, Metro, and even on the radio in Chicago!

None of us in the lab ever expected that this little project would take off the way it did. But it did, and for the next two weeks all development ceased and we were in full-time media mode. Every day we had photo shoots or interviews with another reporter. We even had an entire day set aside to shoot a short documentary piece with the Discovery Channel for Daily Planet (which was possibly one of the neatest things I've ever done).

Discovery Channel documentary piece from Daily Planet on March 6, 2012

The Arrival of Spring

As winter came to a close, we were still had a lot of work to do to get ready for the ICRA conference. Our video had been well-received, and we were selected as a finalist, along with several other teams. But we had a problem: hockey is a winter sport, and our presentation required a live demo of our research. We needed to figure out how to demonstrate our skating gaits inside a convention centre, in May, in Minneapolis.

The obvious solution was to replace the ice skates with inline skates. The basic principles and mechanics are the same; both provide minimal friction front-back, but very high friction side-to-side. The geometry is similar too; both are a long, narrow support area, making balancing on a single foot effectively impossible without some very tricky active balancing tricks.

To convert Jennifer's skates we simply attached some Lego wheels to the side of the ice skate. We have plenty of Lego in the lab -- thanks to a large pile of Lego Mindstorms kits -- so the wheels were easy to find. Lego wheels also have nice, grippy rubber tyres, but roll well, giving us good traction side-to-side, and smooth rolling front-to-back.

Jennifer demonstrating her inline skates in the lab

With some fairly minimal changes to the gait we had Jennifer moving along on a smooth concrete floor. With a few more experiments, some data collection, and some more videos we were ready for the competition.

Competition Day

At ICRA we were the last of about 12 teams to compete. There were lots of very cool and novel entries, including a team from University of Georgia's theatre department who implemented a system to use the robot as an animatronic actor on-stange and a team that made a car out of a Roomba and had the robot driving along a track using visual feeback (an interesting precursor the the DARPA grand challenge). Another team implemented skeletal tracking using a Kinect to allow for tele-presence robot boxing. Yet another implemented autonomous Dance-Dance-Revolution play.

Ultimately though, Jennifer's combination of technical difficulty -- skating is hard! -- and execution took home first-place, and won the lab a forth DARwIn-OP robot, named Jose.

A recording of our presentation at ICRA 2012

Related and Future Work

Winning first-place at the first-ever Humanoid Application Challenge was a boost for the lab's profile. But just because we won didn't mean we stopped working. Our research on skating spawned multiple publications, and opened up some areas of future work.

First off, the problems we left un-solved: Jennifer's leg motors were simply not powerful enough to get her moving quickly enough to sustain a glide on a single skate. Additionally, the problem of balancing on one skate is very difficult, and was never addressed as part of this research.

To address lateral balance on a single skate, I decided to isolate the problem, which spawned the Bongo Board project, which ultimately became the subject of my Master's thesis.

The lab has never officially stopped working on the problem of skating, though my contributions to the field stopped in 2012. We hope to try skating on one of our larger, more powerful humanoid robots in the future. So keep your eye on the sports page to see if we score any more goals on the ice!

Publications

External Articles