'MacGyver' Robot Can Use Found Objects to Solve Problems
Need to MacGyver your way out of a tough spot? Golem Krang, a robot designed by researchers at Georgia Tech, may soon be able to help. A Navy grant is funding work by researchers to create an algorithm that would give the robot the ability to use objects in its environment as tools to solve problems, such as the one pictured in a simulated scenario. (Source: Georgia Tech)
The difference will be that the robot won't be able to violate all kinds of physical laws and limitations. It will never connect a scuba tank to a garden hose to inflate an air matteres to bust open a hatchway. But a robot that was aware enough of it's surroundings to use a pipe as a lever to pry an object off of somebody would be quite an accomplishment. Of course, that would also be a big accomplishment for a whole lot of our population today. The advantages are clear but the level of creative thinking required is beyond most of our population, and probably beyond all programmers, so it will be amazing to see what gets developed.
The downside is that a robot that smart may want to replace us humans. That could be a problem.
@SparkyWatt: I agree that this robot is going to have one big data problem on its hands to both amass and process all of the possible scenarios and data points in order to make any kind of informed decision about what solution to try or how exactly to go about fixing a problem. While I hate being negative about any kind of technology exploration, I'd say this is definitely a "work-in-progress."
It strikes me that this is more a database application than an algorithm. Humans, after all, sort through millions of pieces of data to do the MacGyver thing. From our point of view the concepts:
- A window may be an exit.
- Glass is breakable
- You have to throw a weight to break a window.
- You can't lift more than 45 pounds.
- A chair weighs about 10 pounds.
Are distinct ideas from thousands that we piece together to figure out that we can throw a chair through a window to escape through it. The big problem here isn't going to be coming up with a clever process for figuring this stuff out. It is going to be putting together a massive database of simple common facts that can be quickly integrated into plans.
I sure hope its equipped with Intelligent Headlights. I wouldn't want it blinded by rain or snow as it searches for a "pipe" to lever a smoking HVAC unit off some poor soul. http://www.designnews.com/document.asp?doc_id=247143 All kidding aside, this sounds way, way out these. How is this thing supposed to know just how truly fragile humans are to avoid inadvertanyly killing the trapped guy with that "oh-so-handy-and available" pipe?
GlennA, I have to agree. I didn't believe half of what McGyver did, so I find it even harder to believe a robot can do it. I'm also confused by the specs. The statement "ability to lift 100kg -- the combined weight of its two arms" is confusing. Does this mean the two arms together can lift 100kg and that they also happen to weigh 100kg? I'm familiar with the Schunk LWA 3 (the arms being used on this robot) and I know it can't lift that much (it's more like 6kg).
I think there's a lot of hand waving with the technical capabilites. It'll be interesting to see if they even come close to what's being touted. I suspect most of the work will be in software modelling of the environment and deducing what's relevant to accomplishing a certain task. That alone will consume all $900K. Hope to see great things come from this.
GlennA, I agree completely. Even the name "Golem" is a word that has negative connotations (e.g. dumb or helpless).
It's a nice idea, but I really would NOT like to see an autonomous robot until their cognitive ability is a LOT further along. And then I wouldn't like to see an autonomous robot because I don't know that I would trust it's motiviation (think Stuxnet).
This sounds ambitious to the point of being too far-fetched. I think sometimes these projects are meant to evaluate 'bleeding-edge' technologies, determine the short-comings, and make a wish-list of new technologies.
Focus on Fundamentals consists of 45-minute on-line classes that cover a host of technologies. You learn without leaving the comfort of your desk. All classes are taught by subject-matter experts and all are archived. So if you can't attend live, attend at your convenience.