Subscribe to Choscura's Blog        RSS Feed

Further Thoughts on Cooperative robotics

Icon Leave Comment
I'm writing this partially as a personal notebook, and partially as a key-features list for a cooperative robotics project.

Processing information for a robotic community should be divided up a few different ways: first, there should be information on a personal basis, or information that a single robot needs to know that no other robot needs to know. Second, there should be community information, or information that is of use to a coherent group of robots, but at the same time as absolutely concise as possible in order to limit the number of transmissions needed to convey the info. Last, there should be vital community information, or information which is always passed to the rest of the group.

However, the information in each of these cases should be dependent on the environment's state and the community's state within that environment. Basic needs for survival should be a first priority for a stationary community, but not a mobile one; basic needs for continued operation should take precedent when continued operation is deemed necessary, but not when it isn't. At the bottom level, that's all there is to it, but at a higher level, there needs to be the capability to make the decisions like this about what states will be best for the community as a whole.

In order to construct a true robotic organism, there are a few basic capabilities that need to be added to any existing robotic capability. The first of these, and absolutely most important, is the ability to reproduce: this means that robots will in some way need to be able to construct things, either by creating new things from raw materials or by combining things in order to assemble a component. Existing CNC technology here is a very useful thing to look at, but has the drawbacks of being non-responsive (during the process of running a CNC routine, the CNC head does not respond to external stimuli in any way other than to zero itself, thus maintaining accuracy) and further of requiring intelligent design (ID) as a basis for what it will create and how it will create it. From a robotic perspective, this brings a few things to light: first, a robot <community> that can reproduce must necessarily understand and have a knowledge of the pieces of itself that it is trying to create, why, and how. Further, it could be added that robotic reproduction could be given the goal of most efficient simplicity, that is, the most performance from the fewest number of parts and the lowest number of machining operations to create any given part.

There are three primary methods of creating individual components that are useful to consider here: additive construction, deformation, and material removal. Additive construction can be thought of as working with clay: you start with a base shape and add bits until you have the shape you want. Deformation is slightly more complex, and involves changing the shape of something so that it becomes what you want: an example would be origami, or blacksmithing, since in both of these you start with one thing and by various processes, you change it's shape so that it is what you want it to be. Last, material removal: the analogy for this would be cutting out shapes from a sheet of paper.

Another primary construction consideration is the shape of a given piece. This might sound redundant, but having, for example, a rounded piece means that, for the same example, many copies of that piece could be cut out from a round material of the correct diameter (for example from a pipe wall).

About self-design, there are many considerations: deformity of a material, stiffness, electrical conductivity or insulation, whether a material is water-tight, the color of a material, dampening or vibration qualities, all of these must be considered and ideals in terms of the capability of each piece should also be considered. for example a piece with a heavy motor mounted alongside a delicate sensor would necessarily need a lot of vibration dampening on the mounting point of the motor, but for accurate sensory readings, a solid mounting point for the sensor.

So, thinking logically about this, it seems like there needs to be some kind of evolutionary modelling happening to allow for both hard and soft bodies, and to allow for objects to be bound together, to move on a single or multiple axis in relation to each other. it seems like these could divided into multiple tiers of design, so that at a lower level, a 2-dimensional concept for, say, a new type of limb could be developed, and then that could be absorbed and transformed into 3 dimensions by a more sophisticated processor, and the kinks of the design worked out that way, before details of design and possibly some type of FEM analysis is performed by a higher level processor, and the final design can be prototyped and 'debugged'.

In terms of construction and assembly, there are a few problems with existing machines and techniques that will need to be solved. The first and most pressing matter is 'zeroing', or making sure that every cut is made exactly a specified distance away from a 'zero point'. If you watch nearly any video of any kind of CNC machine on youtube, you will probably see the machine touching a specific point (usually a small circular disc) on itself before moving out to the material and beginning to cut. This poses a very significant problem for any free-moving robots, because they are not rigidly mounted in any way to anything connected to the zero point. A possible solution (which is admittedly very complex) would be to have multiple 'zero points' within a close proximity of each other, which would be very useful for alignment but still have some problems in terms of distance. This could be solved in a monoscopic perspective by using some moderate-level trigonometry to figure out the exact location of the bot by the angle of each zero point, but a far more likely solution would be to use a combination of partial angle-only positioning, in conjunction with a self-positioning system made possible by a real depth-of-field, supplied by polyscopic vision. This allows for both extremely accurate positioning and orientation, and error-checking. The only problem becomes the accurate placement of zero-points, which would have to be done by some kind of rigid-body arrangement.

This is all for now, enjoy if you're reading it, ignore if if you aren't.

0 Comments On This Entry


Trackbacks for this entry [ Trackback URL ]

There are no Trackbacks for this entry

September 2020

2728 29 30   


    Recent Entries

    Search My Blog

    0 user(s) viewing

    0 Guests
    0 member(s)
    0 anonymous member(s)