I've seen a couple Baxter videos now and it seems that the robot works very slowly. I understand that it has built in safety to slow down when it sees that there are people around, but I'd love to see it doing something useful at full speed to see how efficient it is. Surely they've tested that.
Speed isn't the main aim of this. Industrial robotics has an arms race in speed and precision but this is instead a beginning of an arms race in "common sense" abilities, robots that can do the work of humans without specialized programming. It already seems to operate at human worker speed.
It probably depends on the task, but from the videos I've seen Baxter seems an order of magnitude slower than a human worker. It looks like Baxter takes several seconds to manipulate an object, but the workers sorting fruit in this video, for example, are doing several manipulations per second.
I think part of the problem with those demos is that perhaps the robot detects humans in the room and I keep reading from their press releases that they do indeed slow down when humans are near and stop entirely when very close.
This isn't getting enough attention. They'll be releasing an sdk in a few months and hoping to start an "app store" like industry. This could potentially be the Netscape moment for service robotics (robots that take over retail shelf stacking, fast food, parts of nursing). Web/App developers should consider learning computer vision/machine learning and get in on this.
This is more like the Apple II moment in Service Robotics rather than the Netscape moment. Baxter is a really cool robot, his end effectors are more limited so that will constrain his ability to add value. That said, like the Apple II which opened up computing to a whole lot of folks who would not have considered it something "they" could do, I expect Baxter to open up pick-n-place type robotics to folks who would not have considered this level of automation. My 'stress test' for these robots are that they are cheap enough to separate recyclables accurately on a moving conveyor. In order to do that you need a certain threshold of dexterity and frequency response (speed), so things that can do that can do lots of interesting things.
As someone who watches far too much "How It's Made" and other engineering porn, one of the things you start to notice is the amount of 'human glue' logic in production processes.
Whenever there's a weird-shaped widget that needs to be inserted into a die at a funny angle, or some apparently trivial decision needs to be made, there's probably some poor bastard zoned into a hypnotic trance of 12 hour shifts putting tab A into slot B with the mild spice of danger that OSHA-minimum safety guards never really quells completely.
Of those jobs, it seems on casual inspection that a surprisingly high percentage exist because some process engineer has cranked the numbers, and it's cheaper to get a human to do it than put the dev + maint time into an automatic solution.
At USD22k, these things have got to be cheaper than even just wages for a single min-wage worker for a year, you might even throw 10-20k at consultants to customise it for you as well (and if there's a strong SDK/plugins market, plugin+custom is waaaay cheaper than full-custom)
This could end up being the Manufacturing equivalent of a CMS.
"or some apparently trivial decision needs to be made"
There be dragons. Those apparently trivial decisions, even if they are truly trivial for humans, can be Ph.D. subjects, or worse, to have them done by robots.
For example, suppose a batch of wrongly colored parts enters the system. A dumb robotic system probably would happily build red cars with yellow left front doors forever.
I'm actually much more interested in another trend, represented by MasterCAM's RobotMaster software, that allows CNC machinists/programmers to program robots in the way they do machine tools.
It's a very different approach, but it means that the end-user (machinist) has the ability to know/control his whole stack to a far greater degree than the black-box & magic approach that ReThink is taking with Baxter. For industries that pride themselves on optimizing details - I think the RobotMaster is quite interesting.
17 comments
[ 2.8 ms ] story [ 50.2 ms ] threadhttp://www.youtube.com/watch?v=T6dKf8MZ-RM&feature=playe...
Probably not.
Whenever there's a weird-shaped widget that needs to be inserted into a die at a funny angle, or some apparently trivial decision needs to be made, there's probably some poor bastard zoned into a hypnotic trance of 12 hour shifts putting tab A into slot B with the mild spice of danger that OSHA-minimum safety guards never really quells completely.
Of those jobs, it seems on casual inspection that a surprisingly high percentage exist because some process engineer has cranked the numbers, and it's cheaper to get a human to do it than put the dev + maint time into an automatic solution.
At USD22k, these things have got to be cheaper than even just wages for a single min-wage worker for a year, you might even throw 10-20k at consultants to customise it for you as well (and if there's a strong SDK/plugins market, plugin+custom is waaaay cheaper than full-custom)
This could end up being the Manufacturing equivalent of a CMS.
There be dragons. Those apparently trivial decisions, even if they are truly trivial for humans, can be Ph.D. subjects, or worse, to have them done by robots.
For example, suppose a batch of wrongly colored parts enters the system. A dumb robotic system probably would happily build red cars with yellow left front doors forever.
That's how the first line looks to me.
Designing a web site template? Don't do that.
It's a very different approach, but it means that the end-user (machinist) has the ability to know/control his whole stack to a far greater degree than the black-box & magic approach that ReThink is taking with Baxter. For industries that pride themselves on optimizing details - I think the RobotMaster is quite interesting.
http://www.robotmaster.com/