Fast Navigating Guessing Robots

holy_calamity writes "A new navigation technique for robots allows them to make predictions about what's around the corner based on where they've been already. It works well in repetitive environments like office buildings. If this were a Japanese project I'd say it'd be useful for robotic secretaries new on the job, but since it's an American one I suppose it'll be used for automated SWAT teams."

Comments

swat

[mastershake_phd]

but since it's an American one I suppose it'll be used for automated SWAT teams.

Ya last corner terrorist, next corner must be terrorist, come out shooting.

Re:swat

[cp.tar]

If it were possible to rate topics like individual posts, I'd be torn between Insightful, Flamebait and Troll.

Re:swat

[cp.tar]

Actually, I'm not American and I don't like what I see it's turning into.

And even if I'm capable of making the very same comment, it doesn't mean I don't know it would be flamebait.

And since I don't want this to turn into another flamewar on American politics or somesuch crap, I'm shutting up now.

Mhmmm

[tttonyyy]

...sounds like just the excuse I need to place spinning blades around random corners in the office "to fend off any attacking robot overlords".

Quick decision maker

[Centurix]

[Enters maze] ... First corner, bushes, snow ... Second corner, bushes, snow ... Third corner, bushes, snow ... Fourth corner, bushes, snow, Jack Nicholson behind me with an axe
+++NO CARRIER

First Person Shooters

[Ohreally_factor]

I could see this being applied to game technology before it gets applied to law enforcement. This is an interesting approach to an AI (or AI-like) problem. The implementation just happens to be (and is well suited for) robots.

Re:First Person Shooters

[zero_offset]

Why? In a game environment, it's possible -- easy -- for the "AI" to have full and perfect knowledge of the world. Guessing is not necessary.

Expect the unexpected

[Silver Sloth]

From TFA

But the method does have limitations, Lee says: "It works well in indoor environments, but wouldn't be very good in less-repetitive outdoors environments." So maybe a hybrid? Whilst in structured environments expect what you know, otherwise expect the unexpected. No one single answer will ever solve all problems (except 42)

Looks like a twist on behavior-based robotics

[EmbeddedJanitor]

This looks like a variant on behavior-based robotics. Instead of just prioritising behavior on sensed conditions, it also prioritises based on expected conditions.

Currently robots really struggle with making good judgement calls. Behvior-based systems only go so far, perhaps this will go one step further.

Re:Looks like a twist on behavior-based robotics

[timmarhy]

the reason is they lack the ability to put things into context. computers compute, therefor they calculate numbers and stats very well, but the context of a number, that elusive subtle meaning that a number has completely escapes them.

example, say i presented you with the number 42. on here you might associate it with hitch hiker to the galaxay or maybe something else depending on the infinte number of ways i could put it in a sentence.

Re:Looks like a twist on behavior-based robotics

[chandlerc1024]

Except that context can be computed statistically very effectively by computers (only sometimes, of course).

Say I present Google's computers with the number 42. Surprisingly enough it associates it with the Hitchhiker's Guide to the Galaxy reference, much as you or I might. Indeed, if you put in a sentence it will do a surprisingly good job of responding to the context of the number 42, all through computing numbers and statistics.

Re:Looks like a twist on behavior-based robotics

[SnowZero]

No, this is a variant on a common "Simultaneous Localization and Mapping" (SLAM) algorithm that brings in elements similar to Image Completion from the computer graphics community. SLAM is a sensor and sensor interpretation problem, not a behavior/reasoning or action/control problem. While there are probabilistic reasoning systems, the only thing SLAM has in common with them is that it uses conditional probabilities.

If you want to know more, feel free to ask.

Re:Looks like a twist on behavior-based robotics

[Farmer Tim]

I think you've unintentionally summed up the problem with designing an all-purpose self-navigating robot quite well: it's as easy as putting Google's database and processing power in a box on wheels.

Re:Looks like a twist on behavior-based robotics

[chandlerc1024]

Almost, but not quite. The problem is just as you describe it, but the solution isn't to put Google in a box on wheels. Rather, we should be connecting the box on wheels to Google, wherever it is. My computer doesn't have Google's database or processing power, yet it can analyze the context of things through Google's datamining capabilities. I think autonomous robots need to be a bit less autonomous until we figure out the robot part more thoroughly. Shrinking and minimizing size will come with time in a near steady progression, but leaps and bounds could be achieved through networking the box on wheels to Google (and related) sized datamining tools. Just my 2 cents though.

Re:Looks like a twist on behavior-based robotics

[Farmer Tim]

I agree "Google on wheels" is not the solution, and I think part of the problem with robotics is the overwhelming desire to make them fully self-contained, when it's neither necessary or efficient for most situations (that is, anything terrestrial).

All a robot really needs to have built-in is the equivalent of a nervous system and a brain stem with basic (using the word "basic" very loosely) housekeeping functions like communications, avoiding fast moving objects or balance in the case of bipedal robots; things where a 100ms lag could be fatal (network congestion is always a possibility). Higher functions, like environment mapping, task-based decision making and database functions, could easily be remote without introducing a significant delay (I'm assuming a self propelled robot wouldn't be used in precision, high speed manufacturing where timing is critical, as the robots we have are perfectly adequate and would be far cheaper).

What disturbs me is that this would be possible with, essentially, an old HeathKit HERO with a WiFi adaptor (OK, the SONAR would have to be replaced with LIDAR for accurate mapping, but since that data could be sent as a raw stream to the host processor it could actually decrease the need for on-board computing power compared to the 1980's robot, ignoring for the moment the overhead for a TCP stack). As far as I can see, what's really lacking is an extensible command protocol and a willingness to move away from the "Forbidden Planet" design paradigm.

But I'm no expert in this field myself. A bit of small scale animatronics is as close as I get :/

stereotypes are all wrong

[alphamugwump]

If this were a Japanese project I'd say it'd be useful for robotic secretaries new on the job, but since it's an American one I suppose it'll be used for automated SWAT teams.

More likely, you'd have a Japanese robot who is a waitress by day and a combat cyborg by night. And she happens to be a vampire from the future. And she wears a bunny suit. And she's also a suicidal paranoid schizophrenic.

At least, that's what I've learned from watching anime. For God's sake, if you're going to troll, at least try to get your stereotypes right.

But since it's *American*...

[Anonymous Coward]

If this were a Japanese project I'd say it'd be useful for robotic secretaries new on the job, but since it's an American one I suppose it'll be used for automated SWAT teams. Or, as is more likely the case, it was a bunch of American college grads being bored one night and wondering if they could make a robot guess what their lab looked like. I'm not sure why the author had to take a cheap swipe at a nationality under the flimsy guise of a guess as to its functionality. I'm not even sure why the author had to guess at its use in the first place; this is a website for nerds, and frankly, something like this is plain and simply cool.

Naaaaah, it has to be for automating our SWAT teams, because we're a bunch of killcrazy cowboys looking for new ways to blow things up. Um... yee-haw?

Re:But since it's *American*...

[stratjakt]

Don't take it as an insult, but as a compliment.

Using this type of technology for SWAT in a hostage situation could very well save lives.

Using this type of technology to make a "robot secretary" is pretty much a waste of time and effort to create a novelty toy for rich japanese executives.

This

[Colin Smith]

Sounds like the same algorithm most drivers use.

Is this really profound?

[Hammerself]

I don't know much about AI, but is the idea of making predictions based on previous data some kind of breakthrough? I'm assuming this is just an application of some firmly established concepts in AI. When confronted with a redundant or repetative data set, make predictions based on your experiences as to the nature of new elements in that set. I mean, aren't we paying these guys to tell machines how to recognize patterns? Is it news when they teach a machine to recognize patterns?

I'd venture that the purpose of this post is to discuss Terminators, and Japanese robot secretaries, and to hail our coming robot overlords. This is just a guess based on a highly redundant data set I've been analyzing (rather than doing my work).

Sound like Predictive Run Encoding

[ishmalius]

The more you know about the context, and the more you know about the result to a given action, the less information you need from the environment (or from the other side of a communication channel). This is the Holy Grail of information theory and data compression, and it seems as if they are applying its principles here. Higher CPU and better expert programming will likely produce some nice results in the near future.

I dunno about killcrazy...

[trippeh]

I kinda thought the SWAT joke thing was supposed to be an ironically self-deprecating throwaway line, rather than that other sort of a line, the one with the hook in it, that everyone seems to think it is. Actually, thinking about it, after all the criticism American defense forces have come under, maybe a little prickliness from you guys is a good thing. Last thing this thread needs is someone quoting how much of the US GNP goes into ADF funding.

Been there done almost that

[noddyxoi]

In my project called DATMO (Detection And Tracking of Moving Objects) i've made a tracker that followed people that could "guess" where the people would appear next, using an industrial laser scanner, check the video at http://miarn.sourceforge.net/videos/pv3d_peopletra cking_and_scene.avi

Re:Been there done almost that

[Farmer Tim]

That's nothing, I've designed a tracker that followed people and could "guess" where they'd disappear, using an industrial laser cutter.

I'm in desperate need of some new research assistants...

Reinforcement learning

[aquila78]

This almost sounds like some variant of reinforcement learning.. (The bit with confidence scores). Why do they never post real algorithm details :-(

Overkill.

[jpellino]

"I'd say it'd be useful for robotic secretaries new on the job"

As they get chased around the desk by their robot bosses? It's pretty much left, left, left, left... etc...

Re:Or.. This could be an opinoin

[AlHunt]

>post such opinions in public (omg !), disregarding the retarded "we must stick to the middle of the >road" mentality. Moderation is not about
> telling people what you think of their opinion.

The story should be presented without editorial comment, however. After that it's open season.

That said, I guess we'll start using fluffy bunnies to sniff out bombs instead of machines. We wouldn't want to violate the Robot Bill of Rights, eh?

Re:Or.. This could be an opinoin

[zero_offset]

The actual problem is that when these statements are made in the story summary, they are not subject to any moderation, deserving or otherwise. Regardless of whether you feel the statement deserves moderation, it clearly isn't adding anything to the summary. It's an old slashdot problem, and you can bet that the comments which survive just happen to match the slant of the editors: in effect, by making it into an untouchable story summary, it received the ultimate up-mod...