I've been thinking for a while about low-cost electronics projects. What could you make out of the garbage? People throw away electronics all the time, and recycling it is an interesting sustainability and autonomy experiment.
The ultimate goal of Ghettobotics is a self-sustaining industrial economy that consumes nothing but discarded electronics and other trash and produces, with a minimal amount of human effort, useful robots. To this end, every tool and material used in the Ghettobotics program can be made out of trash, by Ghettobotics itself.
A big part of the problem is that you can't make much out of electronics without some fairly specialized tools. So the first problem is how to build the tools out of garbage --- maybe using a previous generation of tools you bought in a store.
So what tools do you need? These would be desirable:
You also need some basic electronic and mechanical parts; the following are desirable:
In taking apart and buying electronics, I'm highly impressed by the degree to which they seem to have been designed by people who can't program, and therefore prefer electronic or even mechanical complexity to simplify software. Now that you can get low-power microcontrollers for under a dollar, this seems absolutely idiotic to me, except in cases where software simply won't cut it. I conclude that knowing how to program is an incredible secret weapon in electronics design.
This is basically a resistive wire that can get hot and has a pointed tip. This should be relatively easy to improvise, I haven't tried. In a pinch you don't even need the resistive part; you can heat up a chunk of metal with fire.
For cleaning the tip, if cellulose sponges are hard to find, you can use wet paper or cardboard.
A temperature-controlled soldering iron would be really useful; you can use a thermal-resistance "voltage divider" with your hand as the known-temperature "ground" to measure the tip temperature with pretty much any old temperature sensor. Then you just need a little hysteresis and something to turn mains power on and off with; a triac? Not sure. A high-power relay would work.
Desoldering parts with just a soldering iron is feasible, but tricky and messy. It's especially difficult using just a low-wattage pencil-type iron, which simply can't handle desoldering high-current or heat-sinking connections; a high-power soldering gun would have less risk of damaging chips. Hot-air heating is apparently the standard approach here for rework stations, but I imagine you could also use a reflow oven or hot plate.
Rework stations also have desoldering irons different-shaped tips for desoldering different parts quickly, heating up all their pads at once. I don't know how to improvise this.
Beyond just heating solder up, it's also handy to have other metals to mix into it to lower its melting point. Standard lead-tin solder alone can help if you're trying to desolder modern lead-free devices; lead might work too. The other alternatives that occur to me sound either expensive or, in the case of mercury, dangerous.
Solder braid or a solder sucker also helps a lot. Presumably if you have some copper wire sitting around you can braid it into solder braid, but a solder sucker would help more if you want to be able to recycle the solder. An improvised solder sucker could maybe be made from a metal tip with heatsink fins on it and a rubber squeeze-bulb, but it's going to be difficult to make something that performs as well as a standard Teflon tip.
The traditional way to build a VOM is with a galvanometer, a network of resistors with a selector switch, and a battery for the ohmmeter bit. You can still do this (hard disks are built around voice-coil actuators that should work well for the galvo) but it's probably more reasonable to use a cheap analog-to-digital converter that measures voltage directly, then use a microcontroller to convert the reading to something comprehensible.
Historically, VOMs have been designed around linear components, because nonlinearity is hard to compensate for in the analog realm. It's really easy to compensate for in the digital realm, so it probably makes more sense to not worry so much about linearity.
The electrically simplest design for a VOM is something like a microcontroller with a built-in ADC, a voltage divider (one large resistor between the voltage input and the ADC pin, and one small resistor from there to ground), a speaker or piezoelectric sounder, and a battery; total of five components. Current inputs connect directly to the ADC pin and ground; voltage inputs go through the large resistor, enabling the measurement of voltages much larger than the microcontroller can deal with directly. An additional input pin with a pullup resistor and ADC input is used to measure resistance. The microcontroller uses speech synthesis to communicate the results of the measurement.
If you're willing to require an external display, you can ditch the speaker and send the measurement results over a serial or USB connection; if you're willing to require an external power supply (say, an unplugged laptop on battery, connected via serial or USB), you can ditch the battery too, getting down to three components, two of which are resistors.
Back-of-the-envelope calculation says that with a 10-bit ADC with selectable 5V and 1.1V range (like what you get on an Arduino) and a 50:1 voltage divider, you can measure up to 250V with 0.25V precision and up to 55V with 0.05V precision. If you then manually connect smaller voltages (below 5V) directly to the ADC pins, you can get up to 5V with 0.005V precision and up to 1.1V with 0.001V precision.
You can use a microcontroller without an ADC if you can convert the quantity you want to measure to a measurement of time to exceed a digital threshold; for example, you can hook up a capacitor between a couple of pins of the microcontroller, instead of a resistor, and measure an RC time constant. You may be able to get better precision this way, but it depends on your capacitor's temperature-constancy and on your oscillator's constancy. You may be able to calibrate to temperature, both by charging the capacitor under microcontroller control at start time, and by using a temperature sensor. (Many microcontrollers have temperature sensors built in.)
A very useful feature of VOMs in general, lacking in the normal ones, is to correct for noisy connections. If you average over time, a shaky connection to a 10-volt power source might look like 9 volts, 5 volts, or 1 volt; but if you look at the voltage waveform at microsecond resolution, you can tell that it's 10 volts with a shaky connection, not 5 volts or whatever. VOMs (and LCR meters) built around ADCs should be intelligent enough to do this correction.
It would be interesting to examine some diode-resistor networks to see if you can get a wider range with good precision by building a nonlinear input circuit, avoiding the need to switch ranges in order to measure large and small voltages --- and, perhaps more importantly, the possibility of blowing up your meter by hooking it up to a too-large input signal. I'm thinking something like using a high-current power-rectifier diode in place of the low-value resistor in the voltage divider, so that the voltage signal you measure is the diode's minimal turn-on voltage drop, plus an additional voltage that's nearly logarithmic in the current that passes through (which is more or less the input voltage E divided by the larger resistance R between the input and the ADC pin).
Something like the ON Semiconductor MBRAF1100T3G Schottky power rectifier diode would be ideal: assuming 25°C, at 20 mA it drops 0.42 volts; at 200 mA it drops 0.54 volts; at 2000 mA, which is above its maximum rating, it drops 0.70 volts; and at 20 A, it drops 0.98 volts. (Serious temperature compensation would be needed to get a reasonably accurate measurement, because the voltage varies by more than a millivolt per °C.) It would be very difficult to drive the voltage across this diode into a range that would damage a microcontroller, even one without 5V-tolerant inputs.
If your ADC precision were 1mV, this would give you measurement precision of around 1% for currents in this range, at the cost of imposing a significant voltage drop.
If we assume that the diode doesn't turn on until 0.42 volts, which is probably about right, then you'd have effectively an ideal voltmeter up to 0.42 volts, and above that, nonlinear response depending on your input resistor. You probably want your input resistor to be quite large, since the diode isn't going to provide much input impedance above half a volt, but you can't make it too large or you won't be able to measure voltage at all. With an absolutely minimal requirement of being able to measure a doubling of input voltage per 1mV increase at the ADC pin, and only trying to measure up to 100V (the maximum reverse-bias rating for this diode), you need 100V to give you 0.428 volts at the ADC pin, which would be about 23mA through the diode, so your input resistor can be at most about 4kΩ. With this network, you can get better voltage precision than the quarter-of-a- digit that this implies, but only at the cost of a lower input impedance.
I have no idea how common Schottky power rectifiers are. I've found lots of power rectifier diodes in the garbage but I have no idea what kind they are. I assume most of them are large standard silicon diodes.
So you might be able to do better with a more elaborate network, or using a separate input pin for voltage measurement. You probably want a diode that's rated for a lot less than an amp. The Vishay SD101A, for example, goes from 200mV at 0.01mA to 1000mV at 15mA, and is rated for 60V max reverse voltage; the same 4kΩ input resistor and 1mV measurement precision at the ADC pin would then give you some ±2% precision on your voltage measurement over that range and down to 50mV, rather than the +100%,-50% you'd get with the power rectifier. That is, with this configuration, your meter can measure voltages covering three orders of magnitude with a consistent ±2% error. That's probably better than the analog multimeter I normally use.
(Incidentally, this configuration might also be useful for digitizing audio and other signals; you only get about 34dB of SNR, but it's a consistent 34dB across 60dB of dynamic range, instead of having a noise floor 34dB below the strongest possible signal. You get 94dB separation between the strongest measurable signal and the weakest measurable signal, as if you were using a 17-bit ADC instead of a 10-bit ADC. Analog Devices recommends this in their application note MT-018, and explains that this resistor-diode configuration is how Bell originally implemented μ-law for voice digitization!)
If you're fishing inductors and capacitors out of the trash, or even just using surface-mount ones, you probably need to be able to measure them; the device you need is an "LCR meter", and they are available new for less than US$40. But variants on the previous section's ohmmeter circuit should be able to provide reasonable measurements.
A really handy physical design for this is the "smart tweezers" design, where the probes for your LCR meter are flexible and elastic, so you can pick up the component under test between them, or even touch them to the component in-circuit before bothering to desolder it.
Many microcontrollers have ADCs but no DACs, which means that stimulating a device under test with a sine wave is out of the question without further hardware. Measuring the impulse response of the component is theoretically sufficient, but you can also generate digital white noise to stimulate the component with, then compute the Fourier transform of the noise and the measured current waveform.
This test provides enough information that you can automatically test the hypothesis that the component you're measuring is actually just an RL or RC circuit, and rather than giving dubious numbers when it's clearly not, you should try some more elaborate hypotheses --- more elaborate network models, or nonlinear components like diodes and transistors.
There are several articles about homebrewing LCR meters with microcontrollers: http://www.kerrywong.com/2010/10/16/avr-lc-meter-with-frequency-measurement/, http://electronics-diy.com/lc_meter.php, http://web.archive.org/web/20080405215220/http://ironbark.bendigo.latrobe.edu.au/~rice/lc/, http://py2wm.qsl.br/LC_meter/LC_meter-e.html, http://reibot.org/2011/07/19/measuring-inductance/. Cypress has an example project using their analog PSoC chips at http://www.cypress.com/?app=forum&id=2492&rID=76890.
This can be usefully separated into data acquisition and display.
At times I've had some success using my sound card in place of an oscilloscope. If the signals are within the audio range, a standard 8Ω speaker can also provide some valuable information about them. But what you really want is to digitize the signal and display it on a screen.
You can, of course, use any ADC. But a standard cheap analog oscilloscope has 20MHz bandwidth, which would require a 40Msps ADC, and actually a 20MHz analog oscilloscope can still detect signals much higher than 20MHz. 40Msps ADCs are not commonplace. AVR ADCs can be run at up to a couple of Msps with reduced resolution (about 6 bits), and I found a three-channel 12-bit 2Msps ADC in a discarded flatbed scanner. Some such scanner ADCs can digitize a single channel at three times their three-channel speed.
Many of the remarks about cheap-shit VOM frontends apply here too. You can use a nonlinear input network and restore linearity in software.
If you want to examine a high-frequency signal that is repetitive or nearly so, you should be able to downconvert it into a lower frequency band in order to digitize it with a lower-data-rate ADC. In fact, if the sample-and-hold circuit on your ADC is fast enough, you may be able to do this with no extra analog circuitry!
Another approach is to use many ADCs in parallel, triggering them in a round-robin fashion. The guts of about 8 scanners would suffice to produce a 40Msps data-acquisition system this way, with a little bit of coordinating circuitry. This, however, also depends on the sample-and-hold circuits being sufficiently fast.
If you have a laptop, you should use that, because you can afford dramatically greater amounts of computation and storage for display that way than if you build it yourself out of garbage. See below.
Analog oscilloscopes provide great variation in intensity, carrying additional information. Oona Räisänen demonstrated the rather impressive difference in her post Rendering PCM with simulated phosphor persistence:
Now how cool is that? It looks like an X-ray of the signal. We can see right away that the beep is roughly a square wave, because there's light on top and bottom of the oscillation envelope but mostly darkness in between. Minute changes in the harmonic content are also visible as interesting banding and ribbons.
For lots of measuring-equipment stuff, you need a display. The traditional approach is to integrate the display into the measuring instrument, but it probably makes more sense to separate them, so that you can amortize the effort of building a good display over several measuring instruments, each of which can then be extremely simple. You may want a simple seven-segment display on the measuring instrument itself (easily recoverable from discarded calculators or watches) but more detailed data display is probably better with a separate display device.
People are throwing out CRTs all over the place, but here in Buenos Aires, it's difficult to find an intact discarded CRT --- cartoneros break off the yoke to recycle, ruining the tube. But if you can find an intact CRT, you can probably drive it with a VGA or composite NTSC or PAL video signal from a microcontroller, without having to worry about any of the electronics inside the case.
The TVout library for AVR microcontrollers works quite well with just two output pins and a couple of resistors to generate a blocky one-bit black-and-white NTSC video signal, which is dramatically better than nothing, but I think you can do better with even a few bits of ADC. A three-bit or four-bit R-2R ladder DAC is easy to construct.
Trickier is that the TVout library uses a lot of memory (kilobytes) for a framebuffer. This dramatically limits the resolution, and inhibits the use of grayscale. More elaborate software could display higher-resolution vector graphics.
A perhaps more interesting approach is to use displays from discarded cellphones, which typically have their own framebuffer. You send them commands to write (and maybe even read) their framebuffer, so you don't need to have enough memory yourself.
Audio output is another, cheaper option. Aside from simple approaches like beeps for continuity testers, there are several implementations of speech synthesis on AVR microcontrollers, one of which is Cantarino (LGPL); I think you can get comprehensible speech at under 100k 8-bit multiply-accumulates per second, which is within the capability of many microcontrollers; it should be about 0.2% of the CPU cycles of an Arduino. Arjo Chakravarty has even done speech recognition on an AVR, called μSpeech.
For testing, you need a bench power supply with selectable output voltage, one that's not necessarily highly efficient, but is hard to burn out. I found an adjustable-voltage DC-DC converter chip in a discarded scanner, but I haven't yet tried using it for anything.
I don't know how to build robust power supplies, but I assume that big power clamping diodes and big current-limiting resistors are involved.
This is especially important for reverse-engineering the undocumented controllers often found in discarded electronics, but you also need it to diagnose problems in the things you build. You can make a slow logic analyzer (say, a few million samples per second) with just a microcontroller using its GPIOs; what remains is only to view the waveforms.
A fast logic analyzer probably needs shift registers or something.
Lots of embedded systems can be put into debug mode and even reprogrammed with a JTAG interface; you can also use JTAG's boundary-scan functionality to map out board connectivity and diagnose connectivity faults in your own devices. Four pins on a microcontroller, hooked up to a standard 6-pin JTAG header, suffices to control other things through JTAG.
One of the most common JTAG cable chips is the FTDI FT2232, which you're not likely to find in the garbage, unless it's garbage from an electronics lab. But you probably want to emulate it in software so you can use existing JTAG software on your laptop.
Felix Domke wrote a paper about JTAG for reverse engineering in 2009, which he presented in much more detail in a talk at 26C3, and the NSA@home project used JTAG to reverse-engineer boards that had FPGAs on them to repurpose the FPGAs for hash cracking; they published software for this called "jrev".
Among other things, JTAG lets you identify JTAG-attached chips automatically and use them as shift registers.
Any old computer will work as a serial terminal, with its own display; but you may want a microcontroller-based serial terminal. A "terminal emulator" program that does what the Arduino's serial monitor does is quite simple; it doesn't even support backspace. A more full-featured serial terminal can easily fit into a microcontroller.
If you're recycling old garbage electronics, their ROMs can be useful for several different purposes:
Dumping the ROM can allow you to repair other instances of the same device that have damaged ROMs, or to reverse-engineer the controller that runs the program found in the ROM.
If the ROM is reprogrammable, you can use it to store other data, or even...
if the ROM is programmable and asynchronous, use the ROM as a programmable logic device whose inputs are the address lines and whose outputs are the data lines.
To reprogram an EPROM or EEPROM you need at least a PROM burner and possibly an EPROM eraser, which is a shortwave ultraviolet light. PROM burners are relatively straightforward to build with a microcontroller, although they do often need control of 12-volt voltage, which you can achieve with many of the chips you'd use for motor control.
If you're building your own circuits, eventually you will benefit from being able to assemble them automatically instead of by hand. I don't know all of what is involved in this, but I think that a lot of the mechanical and electronic difficulties of high-precision X-Y positioning are already dealt with by many inkjet printers.o
This lets you squeegee solder paste onto a printed circuit board so you can solder your surface-mount components all at once with a reflow oven instead of one at a time. It's basically a plotter. You should be able to use overhead-projector transparency film for the stencils themselves, but X-ray film might be an alternative where you don't have sufficient overhead transparency film handy.
I don't know what's involved in actually cutting these. I think a vinyl-cutting machine might work.
A reflow oven bakes your circuit board until the solder paste melts and all of your components are glued into place. SparkFun has a tutorial on their site on how to use a hot plate for this instead, but presumably you can also build your own reflow oven from a metal box, some heating elements, and a thermostat.
Michal Zalewski's "Guerrilla guide to CNC" explains that often it's more reasonable to make even one-off robot parts by cutting molds for them and then casting them than to cut the parts directly. You can go a long way with a three-axis robot for this, which might be the same one that does your pick-and-place.
Long ago I saw a video of an X-Y robot that avoided the difficulties of gantry construction by using two turntables in parallel planes, one of whose edge passes over the center of the other; by this means, you need only rotary motion rather than linear motion. This way, X-Y positioning needs no more mechanical complexity than a couple of bicycle wheels driven by belts from motors, which is considerably less than the mechanical complexity of an inkjet printer.
Recycling solder is a little tricky, but I think a solder sucker is probably sufficient. You probably need to add new flux; I think pine pitch is sufficient, but haven't tried it. You probably also want to keep your RoHS lead-free solder separate from your traditional lead-tin solder.
Ethernet cables typically have nice solid wires that are easy to work with.
It's a pain in the ass to deal with multi-board circuits that are connected by wires soldered to all the boards; they tend to break easily. Connectors are easily recyclable. The most robust connectors are from automotive systems.
It's also valuable to be able to connect to standard cables: audio, video, network, etc. Ethernet cables are especially useful for unintended uses, since they tend to be long.
Small ones of these are all over the place (many of them are literally a dime a dozen new from Digi-Key). The tricky part is keeping them categorized, since you need so many of them. It may be worthwhile to catalog them with an LCR meter in situ, only desoldering them from their original board when you need them after looking in a search engine. Smart tweezers on a pick-and-place robot could theoretically categorize them into little boxes.
Larger items are somewhat harder to find. Discarded old fluorescent light fixtures typically contain large inductors as "ballast"; I've found large resistors in discarded microwave ovens, and of course they're often used as heating elements; large diodes are often found in power supplies. Large transistors, capacitors, and transistors, I have no idea.
As components rather than tools --- I've mentioned CRTs, cellphone displays, and 7-segment displays from calculators and watches, but there are many other premade displays out there.
LCDs can be recovered from broken laptops.
I haven't yet tried to connect new wires to a broken e-ink screen, but if it's feasible, broken Kindles would be a rich source of excellent displays.
Vacuum fluorescent displays are among the most visually appealing to me, but they also require a high-voltage source, so they're tricky to work with.
Laser displays are traditionally done with a couple of galvanometer-driven mirrors, and they look awesome. I've speculated about building laser displays by deflecting mirrors using speaker cones instead, which would make them cheap as dirt; and you could use a laser display with a light sensor as a camera or scanner, too.
These are the easiest components to scavenge: half-broken earphones, dynamic speakers in broken TVs, piezoelectric speakers from greeting cards, and so on.
When I was a kid, we used cardboard boxes for our electronic projects; but nearly anything can be used as a case: empty bleach bottles, books, tin cans, discarded pots, blocks of wood, hunks of styrofoam (modulo ESD issues). Many discarded electronics include particularly robust steel cases, but toys provide plastic cases that are easier to drill.
Inkjet printers typically have two relatively powerful (around 20W) stepper motors and PWM-controlled H-bridges to run them. Flatbed scanners will have one, which will be lower power. Laser printers have motors too. Power-supply fans have brushless DC motors. CD drives need at least two motors, one to spin the CD and one to move the head, but may have a third one to open and close. Disk drives have high-speed motors with the finest bearings in the world. Washing machines, refrigerators, and the like have much higher power motors, into the hundreds of watts and beyond; car starter motors are really high power, into the tens of kilowatts, but only work on a very short duty cycle.
Even small stepper motors probably need more than 5 volts. Chips like the ULN2003 can help with this, but then there are PWM H-bridge chips like the LB1845.
Recent Epson inkjet printers use proprietary motor control chips which seem like they'd be really useful if they can be reverse-engineered.
Bigger motors need things that can control several amps of electricity, often relays.
You need amplifiers for lots of things: driving speakers, controlling motors, and boosting delicate analog signals so they can get to where you digitize them. Amplifiers tend to be fairly specialized. Audio equipment tends to have lots of amplifiers in it, including opamp chips, that won't work much outside the audio range; but an awful lot of useful signals are in the audio range.
ADCs are fundamental to interfacing with the outside world. Some microcontrollers have them built in; if not, you may be able to do delta-sigma DAC with not much more than an analog comparator, such as an opamp, and a resistor and capacitor. But lots of discarded electronics has separate ADCs in it, because keeping the ADC on a separate chip reduces electromagnetic interference. So you can find separate high-quality ADCs in scanners and some audio equipment, at least.
Even more audio equipment has high-quality digital-to-analog converters, and you can cheat by using PWM or PDM at a high frequency; just be careful your high carrier frequency isn't going to blow up any tweeters.
Optical presence sensors are ubiquitous in printers and scanners; the light from an infrared LED is blocked, or not, on its way to a phototransistor. Typically they use them as a "home position" sensor.
Scanners, digital cameras, cellphones with digital cameras, and optical mice contain light sensors. LEDs can also be used as (slow, noise-sensitive) light sensors. (I haven't been able to reproduce the Mitsubishi paper that reported microsecond-range sensing times with reverse-biased LEDs; I get more like hundreds of milliseconds.)
Robotics involves being able to measure how far things have moved and where they are --- at least parts of the robot, and maybe things the robot interacts with. There are lots of ways to do this. Stepper motors emit pulses as you turn their shafts; servomotors typically use integrated potentiometers; optical encoders like those used in ball mice give you a quadrature-encoded signal that tell you how fast things move; optical mice take a high-frame-rate video of the table the mouse is sitting on. I've seen inkjet printers that use linear optical quadrature sensors to sense the printhead position, as well. At the high end, shaft encoders give you an absolute shaft position readout, typically using Gray code.
Digital cameras in general can be used with machine vision algorithms to measure position and motion. This is easier with structured illumination, e.g. a laser pointer.
Binary chain codes are a promising alternative to quadrature encoding to measure absolute positions instead of just relative motion, with some bootstrapping time. Random noise, as long as it's repeatable, should work almost as well as a designed binary chain code.
Accelerometers and gyroscopes are a particularly important kind of position sensor for many purposes: balancing on wheels, flight control, and protection from falls, for example. Off-the-shelf MEMS accelerometers and gyros are what current cellphones and laptops use, but they aren't old enough for many of them to have made their way into the trash stream yet. The traditional kind of accelerometer is a sprung weight controlling a potentiometer slider, and it seems like that sort of thing shouldn't be too hard to improvise, given other kinds of position and distance measurements.
This is one of the biggest problems for actually bootstrapping with garbage. You need something you can program to take advantage of all the amazing digital circuitry you can fish out of the garbage. You can get really awesome Flash microcontrollers from Digi-Key for under a dollar these days, but options in garbage are more limited; garbage electronics tend to be old, and mass-produced microcontrollers often use mask ROM instead of Flash, or have fuses blown that prevent them from being reprogrammed.
One option is to use a regular old computer running a real-time operating system like Linux-CNC's EMC2.
Another option is to find something that has the controller program stored in a separate ROM chip, and replace that ROM chip. This requires you to understand the instruction set of the controller, which is often not documented. Lots of reverse engineering is required here.
It turns out that lots of interesting devices from the last few decades actually do keep their program in a separate chip.
Lots of devices have RAM, though usually small; fewer have Flash. It might make more sense to look for discarded SD cards, which will be hard to notice in garbage, just because they're so tiny; but they're supposedly easy to interface with, because they support SPI. I've encountered a DVD player with a one-megabyte Flash chip in it, and a discarded cellphone with a 1GiB mini-SD card.
Actual old disks might make more sense for heavier robots. A disk that's "only" ten gigabytes is too small to be worth bothering with if you have any money; but ten gigabytes is quite a bit compared to what microcontrollers typically have built in. I'm not completely sure, but I suspect SATA is easier to deal with than parallel IDE/ATA.
Few discarded devices have working batteries. Some have marginal batteries. Fewer have workable rechargeable batteries.
Really building batteries out of trash will probably require re-purifying the materials from dead batteries and building fresh cells.
LEDs are in just about every discarded electronic device. They wear out a bit after long use, but they remain useful. High-brightness, blue, and white LEDs are recent (since about 1990), but you can still find them in trash occasionally.
LCD displays and flatbed scanners typically include tiny cold-cathode tubes and high-voltage power supplies for them.
Occasionally you might find neon lamps and the like. Neon lights last forever.
When fluorescent tube filaments burn out, people typically discard the tubes. But I think it's possible to still run them as cold-cathode tubes; you just need a high-voltage power supply.
You can use capacitive touch sensing to make buttons out of aluminum foil and paint; but also many devices contain multiplexed button matrices. Any light sensor can be used as a button.
PS/2 keyboards are commonly discarded and easy to interface to microcontrollers. Sometimes the discarded keyboards themselves work fine; at other times it is more to your advantage to hook up a couple of the keyboard matrix wires to some other kind of buttons.
I think you can make a resistive touchpad that works like a touchscreen out of pencil lead on paper.
Only occasionally will you find a discarded device with working radio communication, other than cellphones, which are useless except for talking to the cellphone network, but there are DIY options: infrared, ultrasound, 900MHz, 2.4GHz; software-defined radio? Basically any discarded audio or video device (or, here, air conditioner) will have an IR receiver for the remote control, which will also have been discarded --- but possibly separately.
(Recent cellphones also have Wi-Fi, but I'm not seeing them yet in the trash.)
Presumably you could also use the parts of optical presence sensors as infrared transmitters and receivers.
Not only do you need to be able to tell where parts of your robot are, you also need to be able to tell what's nearby.
Aside from the camera approaches mentioned earlier, ultrasound is a common ranging approach. I think you can do ultrasonic ranging with a couple of piezoelectric sounders, but I haven't tried it yet.
Nearly every discarded electronic device has some kind of power supply. Often it's broken, and fixing it or replacing it will fix the entire device. Sometimes it's partly broken, and some of its output voltages work fine. Power-supply chips are versatile and useful for many things.
If you want to reflow solder a surface-mount board, you need solder paste. You aren't going to find solder paste in discarded electronics, and in fact it has a quite limited shelf life. It's also difficult to buy in small quantities. I have no idea how to make solder paste out of garbage.
A lot of things (reflow ovens, temperature-compensated measurement, emergency scram on overheat) benefit from measuring temperature. Essentially every electronic component has significant behavior changes when its temperature changes, so in theory any electronic component can be used to measure the temperature!
Some current microcontrollers have built-in temperature sensors, but if not, you can probably build a bridge circuit with a couple of different kinds of resistors to give you a voltage that changes dramatically with temperature. Carbon-composition and wire-wound resistors have temperature coefficients that are opposite in sign. Alternatively, you could measure the RC constant of an RC constant.
All kinds of high-power devices (microwaves, refrigerators, washing machines, etc.) have relays controlling them. These are handy for robots too. Typical relays are only good up to a few kHz, but reed relays are good up to tens of kHz.
A crucial limitation on microcontrollers in general is the number of available digital GPIO pins. You can use a shift register to turn two or three GPIO pins into an arbitrarily large number of slower GPIO pins. But you aren't going to find a lot of discrete shift-register chips in modern discarded electronics.
JTAG to the rescue! Basically any JTAG-enabled device can be used as a shift register by using JTAG's "boundary scan" functionality. That piece of shit undocumented SoC you ripped out of some scanner, which won't even let you read its program from Flash, will work fine as a 200-bit shift register driven through JTAG!
This is skipping quite a bit of mechanical stuff, but wheels are important for mobility. Wheels are often available from discarded cars or discarded bicycles; bicycle wheels are likely more practical, and back wheels already come with sprockets attached, dramatically reducing the need for gearboxes.
Heavy robots probably need fairly powerful motors to propel their wheels.