Zen, CNC, and Mobile Robots

[ToyMaker]

(rant: ON)
I'm sitting here browsing through the message boards thinking about the questions and answers (and unanswered questions) when it flashes on me (doh! ) that there is a major difference in the design philosophy of cnc robots (which I am learning) and mobile robots (which I know a little bit about).

A typical (hobby) cnc machine might weigh 25-100 pounds and be intended to operate more or less continuously for up to hours at a time. In a mobile robot size and weight are of major concern because they are typically battery driven. The first Honda humanoid robot weighs over 500 lbs and has an operating time of less than 15 minutes (on a full charge ). The first mobile robot I built weighs less than 300 grams and will run for several hours on 4 AA cells.

Another difference that really stands out to me is the software design. CNC software often spews out commands and just assumes that everything is happening as ordered. Mobile robot software relies heavily on sensor input to determine which command to issue and on feedback sensors to assure that the issued command was carried out successfully (and more software to address the cases when commands don't work).

In cnc robots (c'bots) the feedback (when there is any) seems to be mostly linear position (as in x/y/z offset from a reference point) or end of travel, and the sensors are encoders and (limit) switches. Encoders are used on wheeled m'bots for odometry (measuring distance travelled) and to measure joint angle in arms and legs. Potentiometers and switches are used too, but there are many other sensors that are rare or non-existant in c'bots. The more common m'bot sensors include contact, proximity, range, and simple vision systems.

The reason for this disparity in software and sensors is the operating environment. C'bots operate in very controlled conditions where there are (relatively) few unknowns. The start of every operation is setting the reference points. Motion procedes from there in carefully measured increments to completion. No foreign objects (hands, pets ) are allowed to be in the operating volume. The operating environment of the m'bot is often unknown. Even relatively well-known surrounds like a living room floor change in unknowable ways (furniture, toys, pets). This dynamic environment demands a richer sensor set so the m'bot can operate robustly and reliably.
(rant: OFF)

So what? say you. Well, this is to help me get my head around cnc design and implementation . Thank you for your patience.
Comments invited.

robotic regards,

Tom
-------------------
Sit on a hot stove for a minute, it'll seem like an hour. Sit on a porch swing with a pretty girl for hour, it'll seem like a minute. That's relativity.
- - A. Einstein

[Rekd]

Very interesting and informative. I had never thought about the differences in the 2. I guess I won't be using 'robot' when I try to explain to people what a CNC is and what I do with them...

'Rekd teh Compatible: Gracefully accepts erroneous data from any source.

[balsaman]

Hey toy maker...are you interested in cnc so you can use one to build robot parts?

Cnc is a tool. Robot is cool.

Eric

[ToyMaker]

Eric:
Yes, my cnc hobby is to support my robot hobby. My first machine is a pcb router, which will probably do double duty for small 'bot parts (legs, adapters, whatever) until the cookie jar can pay for "bigger, better, faster" .

robotic regards,

Tom

[balsaman]

Do you have any pictures of your robots? I would like to build one some day. More of a model of an industrial robot, using floppy steppers, maybe with a gripper or something on the end. I was planning on using lexan sheet for construction.

Eric

[HuFlungDung]

Somebody needs to invent "pain and fear logic" so computers can learn on their own.

[ToyMaker]

Some people at MIT (Rodney Brooks et. al. in The COG shop) are working on a learning robot. But I think they are using something akin to reward training . When the robot does the "right thing" they tickle its pleasure center ?

robotic regards,

Tom

p.s. there are lots more folks working on self programming computers. google "machine learning", you'll be amazed at the number of hits

T.

[unterhaus]

Anyone that is interested in mobile robots should check out the player/stage software from usc

http://playerstage.sourceforge.net/

It allows you to program your robot(s) and run them in a simulator. It is just like the real world, I work with robots all day and you can't tell the difference, except the simulated ones are much less likely to slam into the wall for no reason and they don't develop bad hard drives. We were trying to develop a cheap robot to use with player, but the guy spearheading the effort left. It still would have cost a coupla thou, but the ones we normally use go for 10-25k USD.

That's one of the reasons that I'm building a machine shop, because I'd like to build robots that are designed by mechanical engineers instead of computer scientists. I'm not impressed by the ones we are using. I hope to buy one like we are using (they are quite popular in the robotic research arena) and make a movie of my robot smashing it to bits.

[ToyMaker]

Simulations are neat visually and can test various algorythms in a number of ways. They put simulated 'bot in a simulated controlled environments, thus skirting some common real world problems (like sleeping dogs or playful cats ).
Simulations also completely ignore some classes of problems that situated (physically in the world) 'bots must contend with, for instance, noisy sensors and mechanical bits that fail completely or degrade less than gracefully over time.
Rodney Brooks (of the COG shop at MIT) and many others build hardware 'bots so that they "don't accidentally skip over the hard parts".

robotic regards,

Tom
= = = = =
Advice for the workplace: "Never believe that you see it all, or see that you believe it all."
- - Susan A. Hodge

[unterhaus]

They should have sensor noise in player/stage, it it's not there already. I haven't been programming robots, I mainly work on the busted ones because nobody else can seem to get them running -- and I try to keep my boss feeling guilty about having me repair them. Unfortunately, nobody around here will pay me to design robots, it's just not economic, and you rarely see anyone in research building robots for the same reason.

I agree that there is no substitute for hardware. The cool thing about player/stage is that player will run a real robot. Stage is the simulator. I don't see the programmers I work with using stage much anymore, but when they were learning, they used it all the time. But we couldn't run our robots without player.
Eric

Originally Posted by ToyMaker Simulations are neat visually and can test various algorythms in a number of ways. They put simulated 'bot in a simulated controlled environments, thus skirting some common real world problems (like sleeping dogs or playful cats ). Simulations also completely ignore some classes of problems that situated (physically in the world) 'bots must contend with, for instance, noisy sensors and mechanical bits that fail completely or degrade less than gracefully over time. Rodney Brooks (of the COG shop at MIT) and many others build hardware 'bots so that they "don't accidentally skip over the hard parts". robotic regards, Tom

[trubleshtr]

I just got 2 new (gently used I.R.B.'s) today for FREE!!!!....both are drven by steppers, they where de-commisioned as they no longer could get spare parts for them (on the controls side). I am hoping to either use the drives for my cnc, or find out if Rutex or another system can handle a 4 axis robot?
If I understand correctly, robots use cartesian type movements? Is there a cheap controls system available that I could drive these things with that anyone knows of???
thanks..

otherwise they will be stripped for parts........

[unterhaus]

One of the hobby type stepper control boards is available in a 4 axis configuration. EMC, www.linuxcnc.org, will run a robot. I assume you are talking about a robot arm?