I Am. When? - Robotics

Herbert 1701 Species D Generations 1 & 2

nospam@example.com (Andrew Maxim) — Wed, 16 Sep 2009 17:01:00 -0700

The previous Herbert robotic life-forms all had a single logic circuit and while some made use of different components, the results were the same from a logic point of view. If the robot has enough power then do something. Not a very exciting logic circuit, but something necessary for all life, even artificial life. We can continue to use this simple logic design in one form or another, even with very advanced life-forms. Slightly modified it can become: If you are hungry then eat. For now we will leave it as is and continue by adding more logic circuitry to the robots.

The simplest logic circuits available are the same as the logic operators taught in any introductory computer class: NOT, AND, OR & XOR. These logic chips can be made to suit the purposes of the next stage in robotic life-form evolution, but would require a lot of additional support circuitry. Lacking space on the demo platform, we will instead opt for an integrated circuit that can accomplish our next task: which direction is the better power source?

To answer this question Herbert 1701 Species D will make use of a comparator chip. In simplest terms, a comparator takes two inputs and determines whether one input is higher than the other. Generally the inputs are voltage levels that are being compared. The comparison between the two voltages usually produces one of two outputs, either a ground level or an open circuit.

For Herbert 1701 Species D Generation 1, the two inputs will be the two variable voltage levels coming from photodiodes (and yes, the schematic shows the photodiodes properly biased for this purpose). This is similar to the variable levels used to "steer" Species C, only the voltage is going into the inputs of the comparator chip instead of the current going to the base of a transistor. The output from the comparator feeds the transistor used to turn a single motor on or off. In order to control both motors, and thus turn left and right towards a brighter light source, a dual comparator is employed with the second pair of inputs flipped from the first pair. The output of the second comparator then controls the other motor.

The dual comparator used is a LM393 variant. I tested out several different manufacturers and models of dual comparator chips using a simple distance test from a dead stop with no power. After three minutes in low light, I measured the distance that the robot platform had travelled. The winning dual comparator is a Texas Instruments TL393, which I had floating around in a parts bin. It is now an obsolete part, but with an inch more travel distance than the next closest chip it seemed worth keeping in the circuit. I did not test the newer LM393 from Texas Instruments, but I would imagine it would be an improvement over the TL393. One surprise was the TL393 outperforming the LM393 from National Semiconductor, which is usually the de facto standard for these types of ICs.

The Herbert 1701 Species D Generation 2 robotic life-form is a slight improvement over generation 1. The only additional circuitry is two more photodiodes, one for each side of the robot. Actually, the two original 180deg photodiodes were moved to point to the sides and the new photodiodes are angled forward. These additional photodiodes provide two benefits: First, the extra diode in series results in lower overall current levels and increased the distance test by nearly two inches. Second, having diodes point at the sides allows for a better comparison of light levels versus diodes only facing forward.

Herbert 1701 Species D

nospam@example.com (Andrew Maxim) — Tue, 01 Sep 2009 17:01:00 -0700

As life would have it, the Maxim Maxim kicked in during my search to find the items needed to create a proper Photovore competition arena. I had figured the 250W halogen bulb would prove the most difficult to find, but it was the first item knocked off my list. The "common items" -- such as wooden dowels or Melamine board -- seem to be outside of my reach; short of paying a hefty shipping cost. Instead, I have decided to move on.

Based upon the tests and competition I was able to perform with the Herbert 1701 Species C robots it was pretty clear that the variable trigger solar engine is the route to go, proving far superior in most tests, particularly low-light and bright-light conditions. Therefore, this will be the species and generation that continues forward. At least for the time being.

Seeing as I have little wish and no money to create new circuit boards for each generation of the Herbert 1701 Species D robots, I have opted to build a simple test platform. While this is nothing fancy -- consisting of a solder-less breadboard, a sheet of plastic, a wheel and some motors -- it will work for the purposes of testing different circuitry configurations, as well as varying components.

As can be seen in the platform images, I have built out the variable solar engine using the Maxim MAX8212 voltage monitor. Throughout this species of Herbert artificial life form I will continue to use this same circuit and will be changing around everything else.

What is that "everything else" exactly? Additional logic, a.k.a. a bigger brain. All the prior generations and species of Herbert robotic life forms have included a single logic circuit. This logic circuit came in the form of the solar engine used to power the Herberts. In a logic diagram it would simply be the question, "Do I have enough power?" Nothing complex, but still a logic circuit. The rest of the previous designs contained no other logic circuitry. Even the light detection for the motors was not a true logic circuit, instead being an analog change that produced a slightly modified outcome.

That all is changing with species D, as I add the next logic circuit to this infant species. Basically Herbert will be going from a robot with a brain of less than a single neuron, to a robot with a brain of, well, still less than a single neuron. But that neuron is growing in ability.

Sometime over this next week I will get the schematics posted, as well as an explanation for what went where and why. Until then, let logic be your guide.

Herbert 1701 Species C Generations 4 - 6 Builds

nospam@example.com (Andrew Maxim) — Fri, 03 Jul 2009 17:01:00 -0700

I want to start off by apologizing for an inaccuracy in the Herbert 1701 Species C schematics. I had grown so accustomed to using reverse biased LEDs as photo sensors that I placed the photodiodes in a reverse biased position in all of the schematics. This, of course, is incorrect as photodiodes function in a forward bias position. The charge circuit for Herbert 1701 Species C Generation 6 is correctly biased as it makes use of an infrared LED instead of a photodiode and should remain reverse biased. All of the affected schematics have been updated to fix this screw-up on my part.

Moving on, I have been building out each of the three species C robotic life forms, generations 4, 5 and 6. Although I will not be labeling each as a separate generation, there are many aspects of the mechanical build that are subject to the same evolutionary process that I have been following for the circuit designs. Use of different components and their placement have just as profound an effect on the effectiveness of each Herbert as the initial circuit design, even more so in some instances. Just as with the trial and error used in those circuit designs, the builds have required much redesign and tweaking.

Beginning the design phase, I had decided on an outer shell to hold the various sensors and solar panel in place. I produced this outer shell using a two piece mold process, which will be used for the depictions in the upcoming tutorial. While this design seemed like a good idea in principle, the application left a lot to be desired. Herbert 1701 Species C Generation 4 was the guinea pig for this design and would have likely yelled at me for my idiocy if it were capable of such.

Utilizing this initial build design, I was able to learn a lot about not only the mechanical aspects of these robotic life forms, but also limitations in the electrical circuitry. The first thing I learned is that mechanical engineering is not my strong suit. The second is that an outer shell becomes too cumbersome for a robot that has limited energy resources. Lastly, in a world where the Herberts would be fighting for the best light source, as in a photovore competition, tactile sensors turn the little fellows into complete wimps. This last part is important, as it will be leading into future generation designs (Yes, the schematics are already finished for such. No, they will not be posted until the other kinks in mechanical design are worked out).

Following the above lessons, Herbert Species C Gen 4 received an overhaul. Not only did I remove the outer shell, but the tactile sensors as well (that rhymes). Additionally, the photodiodes have been moved to the bottom of the circuit board to reduce the amount of blinding, or sensor flooding, that occurs (i.e. too much light hitting each sensor so that it doesn't know which side has more light). The result is build #2 depicted below.

Generation 5 was built out identical to generation 4, with the only difference being the actual circuit design. For Herbert Species C Gen 6, the solar panel required repositioning to the front of Herbert to prevent obstruction of the light level sensor in the charge circuit (i.e. the IR LED). I also darkened the area above the photodiodes using electrical tape and placed a divider between each sensor. The results of generation 6 up to this point follows, with these additional changes flowing back to the other two generations.

There will be additional changes to come in these builds, such as blackening the sides of the directional photo sensors, trying out different photodiodes (I have several different parts to try), and, most importantly, moving the motored wheels all the way forward to provide for better weight distribution. For the purposes of testing each species for the natural selection process, these changes seemed irrelevant and will wait.

I performed several tests with each of the generations. The first was a speed test, similar to a solar roller competition. In full sunlight on a flat level surface, generation 4 won out over the short distance of 1 1/2 feet. Its quick charge rate up to 3 volts and efficient energy design kept Herbert rolling along. Generation 6 was a very close second, despite charging to over 6 volts in full Florida sunlight, this generation will continue to run down to a very low voltage (around 1.5 volts) giving it a slower start but continuous stride throughout.

The second speed test was performed over a slightly longer distance of 3 feet, where the clear winner was generation 6. Generation 5 beat out generation 4 by a slight margin, showing that once it had charged fully the first time, subsequent recharges occurred faster and the extra energy allowed for a quicker distance running Herbert.

The next speed test was performed over a 2 foot length on an uneven surface: the walkway in my backyard. About one foot of the test was on a relatively smooth sidewalk, with the second half moving onto a very uneven stone paver. There was a slight gap, about 1/2", between the two surfaces. Again, generation 6 was the victor in this competition. Generation 4 did not finish as it lacked the power to clear the gap between surfaces and instead stalled out at that point.

The last speed test took place near a window indoors on a rainy day across a smooth surface. This test actual took place quite by accident, but brings home the differences between generations. Herbert 1701 Species C Generation 6 was the only generation to start or finish this competition. In the very low light conditions, not even the low voltage generation 4 was able to build up enough of a charge to get moving, whereas generation 6 happily popped along at a rate of about 1 inch per 3 second interval. On a whim, I performed the same light level test using a photopopper from Solarbotics. As with generations 4 and 5, the photopopper never got moving, even with the very low trigger level of the two Miller Solar Engines.

The last competition is the photovore competion, which I will not be running until all mechanical tweaks have been tested. Just to note, in an actual photovore competition each robot is allowed twice the solar panel surface area as is currently in use (read: two solar panels instead of one), but in testing apples to apples I believe one solar panel will show clear results. I will also be throwing the above mentioned Solarbotics photopopper into the arena to provide four competitors total and a good control robot.

So what did we learn from all this? Number one is that I suck at mechanical design and will continue to tweak out and retest the Herbert Species C critters until I am satisfied. The second thing is that the variable charge level of the Max8212 solar engine seen in generation 6 is superior to the set charge level seen in the other two generations, and even the set charge level of a standard photopopper using Miller Solar Engines. At this point it seems obvious to me which generation should progress forward along this particular branch of the evolutionary cycle that is Herbert 1701. Of course, the photovore competion will be the final word in the natural selection process. Stay tuned.

Robot Sensors

nospam@example.com (Andrew Maxim) — Mon, 18 May 2009 00:11:00 -0700

Spend enough time around robot hobbyists or their message forums and you will come across the two "How Do I" topics that popup over and over again. It depends on the time of year and climate as to which topic is more popular, but the first is "How do I build a flying robot?" To be honest, this question made the mode of transportation for the You Design It project a foregone conclusion before voting even started. Flying is really cool and the number two dream of every man, woman and child, hence the reason so many roboticists want to create a flying robot (the number one involves Rebecca Romijn and the Mystique costume).

That is all well and good (Mmmmm, Mystique), but this entry is more concerned with the second of those questions, "How do I implement robotic vision?" It seems like everyone in the robotic world is obsessed with hooking up a 500 gigapixel camera to their robot and letting their robot see exactly like we humans do. Even more so, they all want object recognition thrown in their as well. This is such a popular request that there are a dozen opensource and inexpensive retail projects out there dedicated to allowing hobbyists to do exactly this. Of course none of these projects ever have the disclaimer that the hobbyist is going to be incredibly disappointed with the outcome, but they will. Oh yes, they will.

In order to see (ha! a pun) why robot hobbyists are going to be disappointed, let us backup a moment and look at the human brain. The human brain is arguably the most complex and powerful logical processor in the known universe (some more than others). Even if you made a silicon processor the same size as the human brain, it would still not compare in power because organic brains are analog processors, not digital (they all lied to you in school when they told you the brain uses binary). In addition, a brain is made up of multiple sections dedicated to performing specific tasks, with one of the larger sections being dedicated to visual processing (striate cortex, prestriate cortex, etc). Basically, a brain is a lot of very powerful analog computers working in parallel and roboticists want to make a single 8-bit 16MHz processor accomplish the same functionality, plus handle all the other sensors, motor control and logic programming. Disappointmentville here we come.

I can fully understand why someone would want to build a flying robot that can see and fully appreciate Rebecca Romijn, but it is not going to happen at the hobby level easily. Throw in a few more processors, reduce the pixel count and make it a 16 color count, and suddenly you are in the realm of possibility. Rebecca is not going to look good at that resolution though, so let’s look at other options instead.

I said in the Herbert 1701 Species C Gen 1 & 2 entry that sensors get skimped on for robots, and to explain what I mean by this I am going to once again jump to biology. We all know the five senses, but most biologists can tell you there are more (and none would fall into the X-Files anywhere). Magnetic field detection is well documented in migratory animals, many snakes (and other animals) have specialized sensors for detecting heat (thermoception), everyone knows bats deal with ultrasonic sound waves, and the list goes on. Robots have all these senses available to them and more, yet rarely will you see more than a couple sensors on a given robot.

I do understand that the organisms people associate with just the five basic senses fall into the "very complex evolved species" classifications. So now it is time to shame that belief with a little more biology. Most single cell bacteria (yes, we are talking micro-organisms here) have both a wider variety of senses and a higher count of sensors than ASIMO, one of the most advanced robots in the world. There are bacteria that are not only covered with touch sensors, but some can even tell you which direction is north from south, know which way is up from down in pitch black liquid while at a zero buoyancy, can sense temperature, know whether there is light or not and how bright it is, and even sense minute chemical changes in the surrounding environment. Single cell organisms. And you want to put two IR sensors on your robot and say that is "enough"?

If the robotic community (hobby and professional) is going to have a hope for making complex robots, we are going to have to loosen up on the sensors a bit. There is a limit to the number of I/O (input/output) ports available on a microprocessor, thus a limit to the number of sensors, but that just screams that maybe you should have more microprocessors to support more sensors. Ugobe understood this a bit when they designed Pleo: more sensors meant more "life-like", which also meant more processors. Granted Ugobe just went belly-up, but that has nothing to do with the sensor count and how much more realistic Pleo was compared to other "toy" robots.

The evolution project artificial robotic life forms are very limited in the number and type of sensors currently, but it is an evolution project. These robots are starting off very simple and evolving into more complex organisms, where, I imagine, the number and variety of sensors will increase with the growth. I am intentionally evolving the Herberts in this manner to increase my own understanding of robotics, and also to generate the best options for each generation. That's my excuse for not having more sensors and processors in each robot (yet), what is yours? Really, when you design a multi-thousand dollar robot (yes, I am talking to you Mr. Universities like MIT, Carnegie Mellon, & Stanford, as well as companies like Honda and everyone who enters the DARPA Grand Challenge), you have given up the right to any excuse.

Herbert 1701 Species C Generations 5 & 6

nospam@example.com (Andrew Maxim) — Mon, 04 May 2009 17:01:00 -0700

The strong survive. That is one of those statements thrown around when talking about evolution or natural selection. It is also one of those statements that people opposed to the idea of evolution warp to mean something other than what was intended. Sort of like a woman slapping a man for shouting out "bare run" as he passes her during a jog through the woods. An ultra-feminist takes the verbal words to mean "nude run", where-as a non-biased person would have understood that the man was shouting a warning about a "bear" and that the woman should "run" as a result. It is why scientists rarely use the phrase "the strong survive" any longer.

Natural selection is a much better term that means the same exact thing. An animal of the same species with one genetic trait is more likely to survive than one with a different genetic trait. Which one survives depends entirely on the environment and the other animals around (including ones from the same species). Take for example two moths; one moth is dark brown, the other light brown. Which moth survives? If the two moths are in a forested area where the tree bark is a dark brown, the first moth is more likely to survive. It does not mean the second moth will die out, just that it is less inclined to survive in its given habitat. If the environment has no predators for the moths, then both moths are equally likely to survive.

Herbert 1701 Species C Generation 5 is an example on this concept. By changing the trigger voltage to a higher value (around 5.6V) we produce a simple adaptation over Generation 4. Up until this point there have been pretty clear reasons behind changes in each generation or species of Herbert. More efficient use of energy, the inclusion of sensors, and the ability to move all have simple logical advantages when implemented correctly (and we covered "correctly" for each as needed). The change in Species C Gen 5 does not provide a definitive advantage over the previous generation, nor is it a definitive disadvantage.

I previously discussed how additional voltage can produce an advantage by offering more power to the motors for stall situations. The disadvantage is that it will generally take longer to reach the trigger point for that higher voltage, and under low light levels that trigger point may never get reached. So which is the winner, a higher or lower voltage trigger point? That is what is unclear.

Were I a gambling man, my money would be on some sort of balance between voltage levels. Even better would be a variable trigger voltage based on the amount of light that Herbert was currently basking in. Herbert 1701 Species C Generation 6 is the embodiment of this concept. Using an IR LED in a reverse bias configuration produces a max8212 solar engine that varies the trigger voltage based on the amount of IR light available. The configuration shown in this schematic produces a trigger level that varies between approximately 2.68V in low light conditions and around 5.7V in direct Florida sun. It is this variable solar engine that is at the heart of Species C Gen 6.

It might seem as if the variable trigger level would provide an advantage over generations 4 and 5, but like the moths, the advantage depends entirely on the environment. The most efficient method of determining advantages or disadvantages for each adaptation would be through nature's very own Natural Selection process. And that is exactly what I intend to do with each of these three generations. The winner of this selection process will be the generation that I will continue to evolve forward, the others will be shelved (temporarily at least).

Not wanting to bias the selection process in anyway, I will not be determining the environment. Instead, the robotics community has already decided upon the environment that they feel provides the best test of a solar robot's (phototropic artificial robotic lifeform's) ability to survive: The Photovore Competition. The competition rules I have opted to use are the BEAM Photovore rules straight from Robogames. Two Herberts enter, one Herbert leaves.

If only I had an audio track of Tina Turner saying that last bit.

Herbert 1701 Species C Generations 3 & 4

nospam@example.com (Andrew Maxim) — Mon, 20 Apr 2009 17:01:00 -0700

Thus far, Herbert has come along a fair ways during the Evolution Project. From a simple solar engine circuit, to a species with sensors and the capability of movement. In the world of robotics this really seems like a simple thing. In the world of evolutionary robotics this is a huge change. An artificial robotic life form that "just is," to one that is capable of self-sustaining behaviors. That really is huge.

The self-sustaining behaviors are still limited in Species C Gen 2. While Herbert moves towards brighter light sources, it will run into problems in the event of shadows or darkness. Once again we wind up with little comatose Herberts.

The reason for this behavior is the nature of the photodiodes in Herbert's circuitry. When there is minimal or no light hitting each photodiode the current flow to the NPN transistor base (ZTX1047A) is negligible, resulting in the transistor not turning on. Effectively Herbert goes to sleep when there is too much of a shadow over its sensors, regardless of how much energy it has in reserve (the capacitor). Not exactly a high survival genetic trait.

Herbert 1701 Species C Generation 3 is the solution to this problem. The addition of resistors in parallel with the photodiodes ensures that current will always flow to the transistor bases. This means that while Herbert has energy, the motors will turn and Herbert will continue on in its never ending quest for brighter light. How much the motors will turn depends on the size of the resistors used: too large of values and there is not enough current, too small of values and the photodiodes are effectively removed from the circuit. It is a balancing act that is determined by the characteristics of the photodiodes. While I am certain there is an electrical formula to determine the proper value, I used the trial and error method to come up with a value of approximately 50k ohms.

The next area of improvement for Herbert is in the form of additional senses. Species C Gen 3 possesses the ability to move toward brighter light sources, but will happily charge headlong into a wall and make a futile attempt to move the wall while expending all its energy.

Combating this overly ambitious and self destructive behavior, Herbert 1701 Species C Generation 4 develops a rudimentary sense of touch. As can be seen in the schematic, this rudimentary sense of touch occurs through the addition of two momentary (normally open) switches. In the robotics world these are generally termed tactile sensors. When one of these tactile sensors is triggered it causes near full current flow to the base of the corresponding NPN transistor, bypassing the photodiodes and, hopefully, causing Herbert to turn away from the object it touched.

A little side note here. When it comes to parts for solar robots, I almost always purchase from Solarbotics. Their prices are fair, their customer service is exceptional, and their quality is generally excellent. They are the premier for solar robotic supplies. As much as I love the company, I hate their omnidirectional tactile sensors. Perhaps it is just me, but I can never assemble these things to work well. And at $4.50 a pair, they are too expensive for me to be screwing up as often as I manage. Instead of trying "to get it right" any longer, I have created my own style of tactile sensor, which is basically the exact reverse of the Solarbotics tactile sensors. I will be posting a tutorial on the creation of these tactile sensors shortly, which I feel are less expensive overall and easier to assemble correctly, each and every time.

Returning back to Herbert, you may have noticed a lack of bread boarding for these two generations. That is because I have begun creating a fully functional artificial robotic life form with this species. This means using etched PCBs and actually soldering in parts. But rather than limit the PCB to a single generation, I have opted to include space for the components of generations four, five and six. So please ignore the through holes and solder pads that contain nothing in the following pictures (ignore the solder job as well, it was the only class I missed in Nuke school).

Should anyone so desire it, the ExpressPCB board layout can be accessed here: Herbert 1701 Species C Generation 4 PCB

If you decide to etch your own board, three circuit boards will fit on the standard RadioShack 2-sided copper PCB board and I have included the ExpressPCB board layout for printing both sides onto transfer film here: Herbert 1701 Species C Generation 4 Double Sided PCB Print Out

Lastly, the electrical component part list can be downloaded here: Herbert 1701 Species C Generation 4 Parts List. All of the connector components are not required if you wished to solder each to the PCB directly (labeled "OPTIONAL" on the sheet). I'm on a budget, so I reuse what I can by using connectors.