Efficient Robots, Part II

I started to respond to a comment on my post about efficient robots, then realized that I was writing a blog entry.

The commenter said that I was simply defining robots differently than Drexler. I disagree. I tried to define robot carefully so that there'd be a gray area of continuity between a really simple robot and a fully flexible general purpose robot. By doing this, and showing that a simple robot can be controlled without computation during manufacture, I can also show that (in theory) a complicated robot can be controlled efficiently. This changes the terms of discussion from whether or not robots can be efficient, to what kinds of robot are most most useful in an elegant nanofactory design.

The machines Drexler showed in his videos, I would not consider robots. They perform only one trajectory to make only one type of product. I'm sure there will be some of these in a practical nanofactory.

On the other hand, a machine that could do any of a dozen fabrication operations (plus, if programmed wrong, a thousand other operations that don't do anything useful), would, I think, be a robot by both Drexler's and my definition. If computation had to be done for each fabrication operation, in order to control that robot, then that robot couldn't be used to do molecular operations on kilogram scales.

My point in these posts is that robots - even very complex ones, the kind that Drexler surely would consider robots - can be sent pre-compiled recipes without computational cost during the manufacturing operation, and thus can (in theory) be efficient enough to do molecular operations.

The trick, by the way, is to build efficient copying and counting operations into the control hardware - the robot controller would be a lot more like a DMA controller than like a CPU.

Chris

Chris Phoenix

Comments

[todd andersen]

It would appear to me the building of “block” diamond would be done by a “simple robot “and moving or blocks by a “less simple robot”. We have talked of this before I recall 14 levels in the MM device from 1 molecule placement to a cube to a larger cube and so on. It would seam to me that a 2-3 axes robotics arm could do all the functions needed. If we are building simple thing i.e. a glass for water, a knife, a simple tool.

I am on the other hand a little unclear as to building a CPU I am having trouble seeing how we place the conducting part of the product into the no conducting part of the device. Point do we build 2 lines one of blocks and one of say carbon nano tubes and at one of the early 14 steps combine the 2 lines. Or are we planning on having many different production lines building “simple” blocks and wires and … then to come together around the 3-4 step of the production of the useful product i.e. the CPU.

todd

[Chris Phoenix]

There's no single right answer for which order to build complex parts in. Note that a CPU doesn't have to be electronic; at the nanoscale, mechanical devices can move at GHz speeds. In Nanosystems, Drexler analyzed a fully mechanical CPU.

It's quite possible that building "blocks" of diamond could be done by a very simple robot, or even without a robot. I'm not saying complex robots have to be used. Just that they can be, if it simplifies the overall design.

[mMerlin]

Looks like I got out to the edge of your gray area of what robot means. From your constraint of control without computation, I got to any machine (operation) that can move items into the correct position, without needing intelligent monitoring, feedback, correction. That gets to anything that can be given a 'push', and has to go to intended location because of existing [physical] constraints: anything that is 'grabbed' will always be in the expected position and orientation, the motion path will always be clear, and when released is either already 'locked' into the desired position, or has no 'path' except to go where wanted. That still allows (zero computational) control, like delaying until the next item is in place to be grabbed, using a [simple] sensor to indicate when an item is in position. That definition would [I think] cover both your lower end simple robots, and the Drexler machines that you do not consider robots. The only distinction seems to be multi-function, or re-configuring the machines to do other tasks. Working with macro-scale analogies, that could be as simple 'mounting' a different tool tip, or changing a cam to get a different motion path.

A possible alternative for complex part order creation (for electronics), is to manage the part supply order, so that the correct part always shows up at the right time to get to the right place. Assuming that the insulator, conductor, semi-conductor, parts are interchangeable as far as the placement machine is concerned. The ordering can be pre-computed, then managed without computer (computational) control. Think of a complex tile mosaic. As long as the 'feed' is in the right order, a blind craftsman can pick and place the (identical to the touch) tiles into a grid to get visually complex results. A simple 'tape' (think DNA strand) can be used to pick the next item to be placed from the correct queue (conditional control without computation). That way the electronic components are no longer 'complex' parts. Very simple regular [crystalline] structure.

[Chris Phoenix]

mMerlin, it sounds like you're describing a different aspect of the machine than I am. You're talking about degree of feedback and sensing. I'm talking about configurability of the machine: whether the same machine can be made to do any of several operations depending on what the designer wants to build.

A robot arm with multiple degrees of freedom is very configurable, even if it has no sensing. A machine that's chock-full of sensors may only be able to do one operation (produce one product).

Changing a cam is a good way to think about how robots can be configurable without requiring computation at run-time. The cam could be amazingly complicated, driving the robot through a multi-step recipe with many degrees of freedom; the robot could have a large number of cams that can be swapped in; and yet, swapping cams is a single operation that can be quite efficient.

The point of my posts is that a long list of digital instructions for a robot is basically the same as a cam. It can be "installed" and the robot can take instruction from it, without computation being required to generate or read it.

It's true that managing part supply order is one way to make a machine generate a range of products - if the operation of the machine does not depend on part supply order.

If the parts are in random-access bins, the robot can be given a string of instructions to tell it which bin to pull the part from next. If the different types of parts require different handling, the robot can be given sequences of instructions suitable for handling each part.

And all of this is possible without incurring computation expense.

Chris

[todd andersen]

I was thinking. I recall the video of the MM device from a few years ago showing conveyer belts moving blocks of carbon by robots placing 1 carbon molecule at a time to the cube. When I looked at the video I ask myself how would the cube be picked up to be added to another cube 9 times larger. The base of the cube perhaps 20/20 molecules is attached or bonded to the conveyer belt. I think we were going to use hydrogen on the top layer of the belt forming a weak bond to the carbon cube above. My concern is how we break the bond and what if 1-2 atoms of hydrogen are left behind there are 400 bonds that need to be broken. Question, what is the weakest bond we can reliably use to hold the carbon on the line as we build the cube.

Then I was thinking of the conveyer belt we could leave the cube on the belt instead of lifting it off after placement of all the atoms for the cube. We change the belt lowering the number of hydrogen atoms holding the cube. If we can move the cube from one belt to another belt with less hydrogen atoms that is empty spaces between the atoms on the new belt with perhaps 300 hydrogen atoms holding the 20/20/20 cube not the 400 on the first production belt. Then another belt with 200 and so on till we have a few well placed atoms holding the cube. We need to perhaps angle the cube on the first belt so as it moves from one belt to another it is not all 20 bonds that break at first, we would have 1 bond brake then 2 then 3 by the time we get to 20 ½ of the cube in on the new belt.

Perhaps all this is worked out I would like to see the paper if anyone knows of it.

Todd

[Chris Phoenix]

Todd, I think you're thinking in the right direction.

IIRC, a few nm^2 of Van der Waals contact area is sufficient to hold a 1-micron cube against reasonable handling acceleration, and a few hundred or thousand nm^2 is enough to hold it stable against bond formation operations.

So it is quite possible to use "surface forces" to hold and manipulate micron-scale or smaller nano-components. The only reason surface forces cause difficulty today is that our tools are so crude.

Chris

[mMerlin]

Chris, agree on describing different aspect of machine, all without runtime computational expense.

Todd, the conveyor belt is / was a convenience for visualization of the functionality. It is not really necessary for the 'belt' to be a smooth surface that would contact the 'carried' parts on one complete face. It could be more like a 'bed of nails' with only enough contact to support the items. Many other variations possible, depending on the properties of the components to be transported, to limit the 'stickiness', or method of making the belt release the part when desired. For example, the belt could be 'peeled' off of the carried part when it [the belt] goes around the 'roller' at the end of its travel. That way, only one 'row' would need to release at a time.