Slashdot Mirror
NASA Boosts AI For Planetary Rovers
transcendent writes "According to Space Daily, NASA is working on increasing the ability of future rover's AI. From the article: 'It now takes the human-robot teams on two worlds several days to achieve each of many individual objectives... A robot equipped with AI, on the other hand, could make an evaluation on the spot, achieve its mission faster and explore more'. Sounds like a good idea, but the article continues, 'Today's technology can make a rover as smart as a cockroach, but the problem is it's an unproven technology'. Another article about autonomous rovers being developed by Carnegie Mellon University is here."
Uh, I sincerely hope that you are trolling!
I don't care what software runs on the rovers embedded computers because:
* these robots won't reproduce.
* the poluting materials (ie rover bodies and lander) are already there.
* current AI techniques might just about make a desicion about whether to take a photo or make a soil sample, given a preprogrammed embodyment, but will *NOT* be creating any novel, intelligent behaviours.
I wouldn't worry about AI just yet.
Autonomy will just cut out the 5 minute lag between action and effect on any data sent to and from mars. Think about that next time you feel like your internet connection is too slow.
Besides, as I mentioned in other post, as we start exploring places farther from the Earth, communication lag will start to get much bigger problem, until finally you'll either have to send humans or AI. And I bet AI, even with some risks associated. will be considerably cheaper, so it's better to plan ahead.
"Two beers or not two beers. That's the question." -- Shakesbeer
I do a Ph.D. in an AI-related field at the moment, and all I can say is: Don't hold your breath. While it is true that AI has made significant progress, a few remarks are in order.
First, the "I" in AI really shouldn't be there. When people talk a diffucult decision problem (e.g. some pattern recognition problem), there comes the point where somebody will say, with a solemn voice: "So, what if we use Neural Networks?" (you can practically hear him pronounce those capitals, while he's creaming his pants at the mere thought of his new awsome intelligent system). People often assume that, because a neural network is a very simple and poor analogy of the brain, that it must have some "intelligence".
Guess what? A neural network is a simple nonlinear function. Period. Training such a thing is nothing more than estimating its parameters by minimizing some (usually quadratic) cost criterion. When you put something in, you merely evaluate a rather simple nonlinear function. There is no intelligence involved!
And then people say: "Yeah, but we have different things as well, such as clustering methods, radial basis function networks, Bayesian (belief) networks, support vector machines, evolutionary algorithms, etc,". They too, do nothing more than estimating parameters (of selecting representative examples) based on the statistics of the problem at hand.
There is a good reason for the fact that "AI" researchers themselves often refer to their field as "machine learning", rather than AI. If anything, I'd call AI "AS", for Applied Statistics, because most of the methods we use are either pure of augmented statistics.
That said, machine learning has achieved some nice things. We can do some simple decision-making, pattern recognition (e.g. face detection) and emulate some limited insect behaviour. There even are some limited commercial applications. But we should be very aware of the fact that most "spectacular" results are merely lab results. I work on face detection myself, and I can tell you that "the real world" (natural photos for me) is a bitch as far as applying methods is concerned.
Support a Europe-related section on Slashdot!
I'm not really concerned with how it handles foreseeable situations (ie. holes in the ground). I have no doubt that the engineers will make it quite intelligent for normal operation.
What I'm worried about is how well it will handle unexpected situations. For instance, what happens both a critical sensor and its backup simultaneously malfunction or if the airbag doesn't get far enough under the lander, and so on. It's these situations that matter, because if the A.I. chooses incorrectly, we could end up with a multi-billion dollar twitching piece of metal on the Martian surface.
-Grym