Slashdot Mirror
Spirit Rover Makes Longest Trip Yet
ivan1011001 writes "Spirit traveled just over 88 feet in an attempt to visit the crater "Bonneville" to look for evidence of water on Mars. Engineers had hoped the rover would travel 164 feet, but Spirit didn't cover the full distance because it spent more time than initially planned studying rocks and soil along the way. This is longer than its earlier PR of 70 feet."
(Time And)Relative Dimensions in space... for the uninformed :-)
:-)
Anyone else think it's sort of funny that you have a probe that travels millions of miles to another planet, and the news is that it's then travelled a further 88 feet
Simon.
Physicists get Hadrons!
Agreed, I'm glad the reason it's moving so slowly is 'so it can do more measurements' instead of 'it broke'.
'The staff in the hand of a wizard may be more than a prop for age,' -Hamá, the doorward
Didn't the Soviet built lunar rovers go much further in a single day back in the early 70's? What sort of over-hyped/overly-specific record is this?
"And the award for longest roving in the past 3 weeks on a neighboring planet by an American robot who's name rhymes with 'kirit' goes to...."
I demand a recount!
"If you want to improve, be content to be thought foolish and stupid." - Epictetus
I don't think it was artificial curiosity - Mission Control gave it instructions to the effect of: "study these rocks, then move towards the crater". They thought it would take x minutes to study the rocks, leaving enough time to travel 164 feet, but instead it took 2x minutes, and the rover only had enough time left to travel 88 feet.
This space intentionally left blank.
I think it is a combination of the two...
Mission control sends a command like:
"Go to Rock A"
"Extend Arm, place payload element X on Rock"
"Let Payload element X analyze rock A"
"Switch to Payload Element Y"
"Let Payload Element Y Analyze Rock A"
(...repeat for each element Mission control wants to use...)
"Stow Arm"
"Navigate at bearing of 110deg until Z time"
Each of the science payloads may take an unknown amount of time to perform it's task - the rock grinder probably moves at different speeds based on the density of the rock.
Also, the driving algorythm probably takes more time to analyze no-so-good paths than good paths.
Sticks and Stones may break my bones, but copyright will always protect me.
Metric units please ! - NASA have enough trouble with Imperial-Metric conversions without the Slashdot breeding another backward Imperial generation.
( Of course, with the pathetic spelling and grammar here, American Literature also seems doomed... ).
Anyone who has tried to go for a walk with a 2 or 3 year old kid knows what I'm talking about. You want to walk, but the annoying little brat will stop and examine very carefully every piece of litter, little stone, gravel or mark on the floor. Half way through the whole thing you'll get tired and just go home.
Exploring that piece of litter, stone, gravel, mark on the floor is the whole point of the walk for a little kid. Ditto for the Mars rovers. Our concepts of what a walk should be like do not apply - there is no predetermined itinerary that must be covered, only wide open eyes that want to understand all the marvels that they see.
Be faithful to your obsessions. Identify them and be faithful to them, let them guide you like a sleepwalker. JG Ballard
That sure sounds like the classic definition of AI as "anything a computer can't do yet". At one point, translation of high-level statements into machine code was considered AI. Then Fortran came along and it's not AI, it's solved, see? Decent speaker-dependent voice recognition was once AI, now it's something you can buy in the store and nobody considers it to be AI.
All the stuff you describe sounds an awful lot like AI to me. Just because it's actually doable with known techniques shouldn't disqualify it.
Mod down posts with a "Free Mac Mini/iPod" sig, they're spam!
Well, that seems to be the 'common' understanding of AI, but in the computer science (and other scientific fields), it has a more specific meaning. Otherwise, factoring large numbers would also be considered AI, although there is nothing intelligent about it, given a good algorithm. Finding that algorithm is what would require intelligence.
Here is a definition I like:
AI is the capacity of a digital computer or computer-controlled robot device to perform tasks commonly associated with the higher intellectual processes characteristic of humans, such as the ability to reason, discover meaning, generalize, or learn from past experience. The term is also frequently applied to that branch of computer science concerned with the development of systems endowed with such capabilities. --- Herbert A. Simon, Professor of Computer Science and Psychology, Carnegie Mellon University
I am nitpicking here, but given an algorithm to extract edges and corners from two images, using the camera calibration values to calculate distance, and creating a map based on these data does not require intelligence, and as such isn't strictly AI.
The robot still follows strict instructions which find the optimal path. It will not learn if this algorithm fails a certain number of times, it will not generalise to make future computation quicker, like a human would. It does not have a concept of the obstacles. It does not get more proficient after doing the same for a while. So, even though it's a brilliant example of applied computer vision and autonomous navigation, there is very little of what is considered AI involved. Hope this clears it up a bit.
Sounds like some of my geological field traverses.
A couple of times, I remember the very strong motivation to stop looking at the rocks when I heard the gentle "thwup-thwup-thwup" of the helicopter arriving to pick me up.
Spirit is stuck there, so it may as well take its time puttering around and getting a good look, though it would be cool if it was able to eventually get to the crater to the NE, and then the distant hills (it is unlikely to get this far, but it would be really interesting). It is one of those tradeoffs that would be familiar to any geologist -- do you expend your time looking at what is right in front of you, or do you move on to complete the traverse to new sites that may or may not be more interesting? It is tricky.
You can see some of this balance playing out in the difference between Spirit's progress and Oppportunity's -- Opportunity it taking its time moving only a short distance and intensely analysing the immediate landing site because it landed right in the middle of some great bedrock outcrops in a crater. With luck like that, there is no strong motivation to move on until the the landing site is thoroughly studied. By contrast, if it had landed out in the plain, it would have been moving off to the nearest crater by now, and probably giving Spirit a good competition (smoother == faster, I suspect).
seems the hype on this thing is way out of scale. I am not trying to marginalize NASA accomplishments, though I do find some statements pretty funny.
When they cut into one of the rocks:
"It went deeper than we ever imagined!" (Few millimeters)
Assessing the landing site:
"We can't believe our luck!" (Flat, with a few rocks)
etc.
Now, I think the rover is cool, and want the science just as much as anyone else does, but the statements from the scientists (or their PR person) are just giddy.
Blogging because I can...
I've seen no proof that the Rover reasons, discovers meaning, generalizes or learns from past experience. It just runs a program which gives it very direct commands.
It's a bit like this: If I ask you to get some good tomatoes, you would break this up into several steps: Go to the market, find the tomatoes, then select some good ones. But what is a 'good' tomato? You will have to rely on your experience, your taste, and the past input from others to determine what a good tomato is. Then you would choose the tomatoes which best fit the ideal you have in your mind.
A computer cannot do that. It has no concept of what a tomato is. It doesn't deduce properties from past experiences. You can program a robot to go to the market (by giving it specific instructions on how to do that), then have it pick up tomatoes which have a certain height, weight, a given hue, and a softness, all expressed in measurable units. The robot would bring back some 'good' tomatoes according to these requirements, but it wouldn't be doing anything remotely intelligent, even though it might look like it from the outside.
Now, an AI approach to this would be to model a tomato internally, for example, using a Bayes net of different fruits, associated with different properties. A tomato would be grouped with similar fruits according to some characteristics. The computer would learn through repeated observations (like a human does), and propagate its deductions throughout the net. For example, a squashed tomato and a squashed pepper are both 'bad' fruit/vegetables, and a red pepper and a red tomato would both be 'good', but a green pepper can be good, while a green tomato cannot. The network gets updated to accommodate these observations and build a better model, up to the point where the computer can pick the 'good' fruit without being told exactly what it is.
See, in the second example, there is learning, there is deduction, and there is reasoning, as well as generalization. In the first one, there isn't. That is the fundamental difference.