Slashdot Mirror
The DARPA Robotics Challenge Was a Bust; Let's Try Again
malachiorion writes: The DARPA Robotics Challenge, the biggest and most well-funded international robotics competition in years, was a failure. After years of grueling work on the part of brilliant roboticists around the world, and millions in funding from the Pentagon, the finals came and went with little to no coverage from the mainstream media. The only takeaway, for those who aren't extremely dialed into robotics, is that a ton of robots fell down in funny ways. There were winners, but considering how downgraded the tasks were, compared to the ones initially announced in 2012, it was closer to the first DARPA Grand Challenge, where none of the robot cars finished, than the Urban Challenge, which kicked off the race to build deployable driverless cars. So just as DARPA regrouped after that first fizzle of a race, here's my argument for Popular Science: It's time to do it again, and make falling, and getting up, mandatory.
>> whole team was not able to attend because they couldn't get their visa in time
Next time, just fly into Tamaulipas and hop on a northbound fruit truck. Visa shmisa.
>> ...expecting smart things out of a people they intentionally made stupid over generations...application of mind control...
Huh? The team that couldn't get in was the Chinese (http://www.bbc.com/news/technology-33045713), and likely as not there was some kind of Chinese/US diplomatic pissing contest going on behind the scenes that WAS on everyone's minds.
While we're at it, notice that my link is from the BBC - if that's not mainstream enough for the OP, I don't know what is.
It is too harsh to call it a failure. After watching many hours of video footage, I would judge it most certainly NOT a failure, but a good first step. Yes, many robots fell down. Yes, it would be nice to make a requirement for them to get up -- and at least one did in the *competition*, if i recall correctly. The tiny robot an Asian student was *demonstrating* (he didn't speak much English) got up amazingly fast since he designed it in.
It is understandable that DARPA reduced the difficulty in this first baby step of the competition. What would you rather have, (a) very difficult tasks so that no team can complete all tasks, or (b) difficult enough so not *all* teams can complete all tasks, but some can? I'd choose (b) every time, since it results in encouragement to take the next step in the development. There are many other benefits if you think about it for a minute or two.
Asians are pulling out of US economy, or rather no longer want to purchase US debt, kind of a heated situation. There is a multitude of reasons behind this. Referencing stupid is for 3 years we couldn't get a robot to remain standing even under downgraded tasks. Perhaps they are being 'punished' because they weren't being 'good consumers'? Doubt it. This is a ruling class management malfunction that put the US where it is today as a whole, and don't look now but the 1%er's are deserting the ship while their corrupted and patsy politicians have international arrest warrants being issued. It is what it is.
There were robots in the competition that did just that.
Walking wasn't a requirement.
Slashdot social media options: AIM, ICQ, Yahoo, Jabber and Mobile Text. Why no MySpace?
... and plenty of the crawling robots also ended up falling over.
But why no love for the videos of robots failing and falling? There are plenty of videos of legged robots not falling, and they are positively terrifying for the humans vs. robots crowd:
https://www.youtube.com/watch?...
Moo?
APK likes to ask for responses to the same things over and over. Maybe he just likes the responses?
... until at least sometime after the Year 3000. We already have the video evidence that robots of the future struggle to get up when placed on their backs (https://www.youtube.com/watch?v=1sNMHdaJx5c&feature=youtu.be&t=5m17s).
I knew I needed to stop reading Slashdot and finish my PhD when I started to miss articles by Bennett Haselton.
I went for one morning of the competition, and there was a LOT of standing around and waiting time. Even when the robots were in action, they would take very long pauses in the midst of the narrowly defined activity they were performing.(and yes, sometimes simply falling over during the pauses). It would have been hard to pick more than a few seconds of footage that would have made it into the news.(mostly the falling over clips)
Walking is not one of the necessary things for robots to do.
Crawling more than suffices. In fact, having wheels is a feature, not a bug.
I would go one step further and at this point split the competition into two parts. Step 1 should be to have a remote controlled robot that can complete a series of complex tasks. At this point why should a robot have to be autonomous and work with degraded communication? The goal is to go into places like a nuclear meltdown where a human can't go. I see no reason why it can't be controlled remotely by a competent human. Get the platform down, get it working great, get it rock solid, then add the AI to a platform that already works great with a human controller. I see no reason to try to solve 2 separate complex problems at once. Just solving the first problem would benefit everyone greatly and make the 2nd problem much easier not to mention that mechanical engineering and AI programming are different domains with different experts and you could let the different experts work on what they are good at.
Exactly! That is how you go about solving engineering problems. Get the fundamentals working well first. It's like trying trigonometry while struggling with arithmetic. Yes, you can do it, but it's limiting with fragile results.
It's not a solved problem or they wouldn't be having all the robots fall down. And if it IS a solved problem then give every team a good robot to program (or a robot simulator if the physical robot is too expensive). The point is to solve one problem at a time. We know how to make cars, bulldozers, etc... Make the autonomous robot control one of those, don't try to add intelligence to untested and unreliable hardware, get the hardware working right first. If a human can't do that task via wire then you're asking a robot to basically be more intelligence than a person. I would say the same of most other AI projects that I see. If a human can't manually control the robot to fold towels in near real time then why should we expect an AI to be able to do it?
...let's not and say we did.
systemd is Roko's Basilisk.
Watching the "compilation of robots falling down" video was quite uncomfortable, because many of them reminded me of my father, who has Parkinson's Disease. I wonder if there is anything to be learned from the similarity. Probably not.
In fact, having wheels is a feature, not a bug.
Until you encounter a steep set of stairs, then simple wheels might not be sufficient. You'll either have to develop special wheels or tracks - which might not be that great at other non-stair-like terrain - or try a more universal mode of locomotion.
Ever see those photos of mountain goats climbing the nearly sheer face of a dam? I'd love to see a wheeled robot do that without grappling lines!
=Smidge=
Moo!
APK likes to ask for responses to the same things over and over. Maybe he just likes the responses?