South Koreans triumph in US robot challenge | Hacker News Reader

South Koreans triumph in US robot challenge

phys.org

95 pointsdnetesn20 comments

Fast motion video of the winning robot: https://www.youtube.com/watch?v=q_yfvlplJe0

1min trailer of the winning robot: https://www.youtube.com/watch?v=L4B5BhDoS9o

That was a big change in the standings since yesterday. IHMC and KAIST weren't even in the top 3 as of 5 PM yesterday. I thought Tartan Rescue had won.

I get the feeling that nobody responding here watched the competition. Is there a forum where anyone is discussing this subject?

Notes:

- Everybody using the Atlas robot had stability problems. Nobody seemed to be outside of static stability (base of support under CG) let alone outside of ZMP stability (base of support under CG modified to account for momentum.) Any loss of balance meant a fall. They're probably using the basic DLL that Boston Dynamics provided to get teams going. (One team admitted this.) That has functions "walk slowly", and "stand maintaining balance using only joints from the hip and below", plus direct control of joints. Boston Dynamics has demonstrated much better locomotion control than that, but the teams didn't have access to it. (Boston Dynamics supplied the Atlas robots and thus could not compete.) IHMC was working on balance recovery [4] but didn't have it working in time.

- This is the new Atlas robot, unveiled in January 2015. It's still mostly hydraulic, but it's now self-powered, with a big battery backpack. I think there's also a hydraulic accumulator in there, so they can recover some energy when running, but nobody at the event went fast enough to gain from that. Mechanically, that machine is far more capable than it appeared to be in the competition. It's quite capable of picking up objects from the ground, and getting up if it falls. Nobody did either in the competition. The teams only got these new robots around February, so they really haven't had the time to develop software to use them effectively. Right now, the hardware is much better than the software.

- Everybody was doing their manipulation by almost pure teleoperation. As soon as DARPA introduced delays into the data links (which they do to penalize teleoperation), the robots spent most of their time stopped. The operators were stuck waiting for enough data to get through to update their screens. Not much autonomous operation here. Here's the winning team video, sped up about 30x.[2] Note that although it's capable of walking, it was run in kneeling wheel mode almost all the time. Only for the stair climbing did it walk.

- Nobody had combined manipulation and locomotion working at all. That's why getting out of the vehicle was so hard - it takes coordinated leg and arm movements.

- The CMU team had a partially tracked base, and the NASA/JPL team had a "monkey" robot with rollers at knees and elbows. They both did reasonably well. If you're going to teleoperate, a dumber, statically stable robot is easier to deal with. Of the top 3 finishers, only one used an non-wheeled robot, an Atlas. Here's their first run, with two falls.[3] Again, this is a 30x speedup. The penalty for a fall or a need for repair was only a 10 minute penalty, not disqualification.

- The tasks were dumbed down twice during the project. The original specs called for the robots to have to replace a pump. The final competition only required turning a valve and cutting a hole in very soft material with a power drill. Basically, the tasks were reduced to the point that they were hard, but not so hard as to lead to total fail.

- Schaft did better in the first round back in 2013, but after Google bought them, they were pulled from the competition. Google, remember, now owns Schaft, Boston Dynamics, and several other robotics companies.

I look forward to the next competition. If work continues, in a year or two the results should be far better.

[1] https://www.youtube.com/watch?v=27HkxMo6qK0 [2] https://www.youtube.com/watch?v=q_yfvlplJe0 [3] https://www.youtube.com/watch?v=vDXnlgNkRl4 [4] https://www.youtube.com/watch?v=soPM7LddLY8

This is all more or less correct… just a few things to add (I'm on Team IHMC Robotics) :)

1. It's highly unlikely that this version of Atlas can get up from a fall under its own power, at least not expediently; it weighs much more than the old one and it actually has a slightly weaker hydraulic system. The old pump motor was a constant speed/constant displacement 3000 PSI system. The current one is a variable speed/constant displacement system that tops off at around ~2600 PSI, and the robot is around 35-40 lbs heavier than the old one. Even if the arms are strong enough, none of the COTS grippers that integrate with the robot could handle getting up; the fingers would explode. I've seen lots of videos of the old tethered Atlas getting up. I find it amusing to think that the current one could pick itself up.

2. We (IHMC) don't do ZMP based stability. All of our walking/balancing is based on something called Instantaneous Capture Point, that the head of our lab pioneered during his PhD and which several other researchers have gone on to refine over the last 10 years or so. Lots of people use it for fall recovery (MIT does in their recent Atlas video that's similar to ours) but we also use it for walking as well. We call it ICP, but vision people get cranky when we say that because ICP is also short-hand for Iterative Closest Point.

3. amjaeger is correct, our robot was very badly damaged by our falls on Friday. Enough that we were very worried about how it would impact our performance. Our falls on Friday were both more or less the result of operator error (with some hardware problems mixed in, but mostly operator error), and our stability issues on Saturday were the result of the falls on Friday.

4. The comment about manipulation being pure teleop is a little reductionist. It is teleop in the sense that it's not autonomy, but almost nobody was doing direct joint servoing which is what I usually think of when somebody says pure teleop. For example, we actually operated continuously during comms blackouts. When we cut out the wall on Saturday, almost the whole thing happened during a blackout. But when your 400lb robot doesn't have a safety line, it behooves you to be extra careful. We cut out the wall just as slowly during perfect comms back in the lab because there's lots of ways to screw that task up.

I didn't follow the contest seriously. Mostly I was just rooting for JPL's Robotsimian since I liked it's design. In addition to stability the wheels KAIST and CHIMP was able to just plow through the debris near the end.

I'd been joking with my co-workers earlier that we should enter BEAR[1]. Of course we haven't used it since I've been at Vecna, we wouldn't actually stand much of a chance of winning, and we've got commercial robots to work on.

[1]https://www.youtube.com/watch?v=_bpd_ycg4hY sorry it's a bit over the top

Hi, I was there! I was on the field team WPI-CMU's robot Warner. (I am one of 4 undergraduate students on the team, I have worked with the team for 1.5 years. I worked on UI's for operator control, robot testing and many other side projects). Just thought I would mention that our Atlas Robot never fell down or needed a reset. We completed 7/8 tasks, we unfortunately failed the drill task twice. The first time because of software error, and the second time a hardware error. There will be some papers coming out in the future from my team, and I am sure there will be plenty from the other teams. I am using in flight wifi, so I will not be providing youtube links to my comments, but I encourage you to check out the youtube pages for WPI-CMU Warner, MIT, and IHMC

To respond to some of your other comments: WPI-CMU, MIT, and IHMC all had their own balance controllers and specialized walking for atlas. I do not know as much about the other atlas teams. The atlas robot is ~400 lbs and ~7 feet tall. My team looked into fall recovery, but we found that the arms we not capable of doing a "push-up" to get the robot off the ground. At that point we decided the best strategy was to do everything in our power to avoid a fall. My understanding is that both MIT and IHMC damaged their robot during their falls. Any attempt to recover would have likely failed anyway. If a fall occurred the safest option was to e-stop the robot to turn off the hydraulic pump. Picking objects up off the floor is tricky. Our team has done it successfully on occasion during tests, but it was very risky and also often caused the robot to fall.

In many cases the software reaches the limit of the mechanical capabilities of the Atlas robot. Even though the Atlas is the cutting edge in robotics getting out of a car is a challenge because it's range of motion is incredibly limited compared to that of a human. It takes Warner ~3.5 minutes to get out of the car and onto a platform we added on. This the result of 2 years of work by a brilliant researcher who exclusively worked on egress. It might be true that most manipulation was pure teleoperation for some teams, but that is not the case for WPI-CMU. I also think MIT and IHMC were highly automated. The pauses in operation were often the robot running planning algorithms. For example we click a point on the operator screen to select an object - such as the drill. Then the robot autonomously determines the location, and finds the most efficient way to grasp it using a technique called trajopt. (If I'm not mistaken) Tartan Rescue used another technique called RRT's - it's rapidly-exploring random tree. These types of algorithms take time. Regardless, after some practice operation with full communication and with com dropouts had similar run times.

Manipulation and locomotion are difficult because manipulation changes the location of the center of mass on the robot. WPI-CMU uses a static stepping technique the always shifts the center of mass above a stationary leg, it then lifted and moved it's other leg, then placed it down. The robot would shift it's weight after the foot is on the ground. Moving the arms while one foot is in the air is not easy. IHMC has a karate kid video where they balance on one foot and do a few kicks and move their arm. It is very very impressive, but not easy.

Originally teams were given 30 minutes to complete each task. Now teams had 1 hour to do all tasks. Things were going to have to change. Hopefully everything I wrote made sense, and helped clarify things.

To the South Korean team: Congrats!

This is just the beginning, both for robots and for asian dominance. Culture plays a huge role, and in our culture we're happy to watch Reality TV and let government babysit our kids. Government doesn't care about robot competitions. I assume in Korea but believe in general in asia, the parents are more on top of their kids, so that even if they have a "lower quality" government (Which is always debatable) in China, for instance, eventually we're going to be less and less the source of these kinds of technical advances over time.

Hell, even if the chinese culture doesn't inspire creativity and achievement, just by demographics they should produce three times as many innovations as us, since if the innovative people are %0.1 of the population, they have three times the population.

Whenever I read Western diatribes about China now, I can only think of http://www.washingtonpost.com/opinions/for-american-pundits-...

"Government doesn't care about robot competitions."

That was a DARPA competition they won.

Which confuses me. I wasn't aware DARPA opened it's competitions to non-USA teams.

Not a big deal. Do a little research instead of making assumptions. I found a German entry in the 2007 Grand Challenge:

http://en.m.wikipedia.org/wiki/DARPA_Grand_Challenge_(2007)#...

Well, now you're aware.

Unfortunately, it seems my comment was taken with an assumed connotation. I simply wasn't aware that DARPA opened it to any one outside of the US since, well, it's DARPA. It probably wasn't the most insightful comment in the world, but, yes--now I am aware.

why wouldn't they?

D is for Defense

Your entire premise is wrong. Japan has proven how easily an Asian giant can fall (just as easily as any other), how quickly they can lose their edge and stagnate.

Demographics and the size of the population has absolutely no correlation to production in any regard.

Pakistan has 182 million people; Nigeria has 173 million people. By your incorrect premise, Pakistan should produce ~55% as many innovations as the US, and Nigeria ~52%. Whoops, turns out it's a lot more complicated than having a lot of people.

Your comment is just speculation.

Currently the most nobel prize won and research papers published is USA.

There are many areas that one can dominate, I'm unsure that Robotic is Asia as a whole continent and also when you mention Asia... Middle East and Russia can be consider Asia too. I'm not sure what you are comparing this to probably USA or Western countries but I don't think this comparison is useful at all cause of the generality.

But this is just glass ball looking.

Good job for Korea and I think it's good for the whole world if more and more countries gets better at things.

Yes, parents do push kids quite hard in South Korea, because studying hard and getting into good schools are seen as plain necessary for success in life. But that doesn't necessarily mean the quality of education is fantastic -- a lot of the after-school classes kids get worked to death in are snake oil levels bad, and the teaching methods applied in the primary schools are often sub-par as well. Meanwhile many students suffer a great deal in that highly competitive pressure cooker, and it contributes to an unusually high teenage suicide rate for the country.

Education and parenting are complex. Education is not as simple as quantity-does-the-trick, and parenting isn't as simple as applying pressure.

Why is it that after looking at this rather cool competition, and looking at all the cool technology that is coming out of it, you reduce it to "South Korea > USA?"

Reply on news.ycombinator.com