Wednesday, April 30, 2014

Automated Investigator

It's time for a new project post. What have I been doing since I finished the LED Planet Project last month? Mostly the things I actually get paid to do, but also a decent amount of slacking off. Without a project that has been handed to me by someone else as a challenge, I've had time to consider what types of projects I want to do for this blog. A while back, I designed and programmed my own flight controller for a hexacopter platform, so I guess I could say something about that. Unfortunately, that was a fairly lengthy project, so organizing my thoughts on the subject will take some time. I'd also like to do a few new experiments with flight controllers, which will add to the complexity of the write-up.

So while I've been trudging along producing interesting content about flight, I figured I could use a little weekend project to have some fun. Just like my project from last November where I used an accelerometer to measure muscle activation, I decided to limit my supplies to only the hardware I had sitting around my apartment. At the very least, this requirement would help reduce the electronic clutter I've accumulated.

As for the goal of the project, I took a suggestion from my brother. He was brainstorming for an engineering project in college that needed to use physical sensing coupled with advanced computation on a computer. He had the idea of making a lie detector (polygraph), where a set of sensors monitor a human subject and a computer determines whether or not the subject is lying based on the sensor measurements. I decided this was an idea worth stealing. So I chose to make a lie detector.

Now, I don't want you thinking that I would try to one-up my own brother in his college engineering project just because I have the resources and free time. My plan was to boil his idea down to something much simpler yet still retaining the core concept, then post my work on the internet to become famous. The polygraph was invented in the 1920's, so neither of us are doing anything particularly original anyways. But if the last two years of internet trends tells me anything, it's that original content is not required to get pageviews.

Sensing

How does a polygraph work? The core concept is that a subject under interrogation will have some sort of uncontrollable bodily response when lying. The validity of this statement is debated in the scientific community, but we won't let a bunch of boring scientists stop us here. If the subject believes that there will be some kind of negative repercussion if they to the investigator, their body is likely to act differently when telling the truth or lying. The reactions commonly associated with lying are an increased heart rate, sweating, changes in breathing pattern, eyes moving in a particular direction, etc. A polygraph will measure one or more of these effects in order to determine whether or not the subject has just told a lie.

Heart rate sensors are easy to come by, and almost as easy to make. The transmission of infrared light through your finger or earlobe is modified by the amount of blood being pushed through. If you were to shine infrared light from one side of your finger to an infrared detector placed on the other side, the detector would see a series of pulses in the amount of light detected. The rate of these pulses is your heart rate. Unfortunately, I didn't have any infrared LEDs sitting around, so I didn't use heart rate sensing for my lie detector.

Your body is a decent conductor of electricity, as long as you can get past the skin. Sweat is a good conductor, and can help reduce the electrical resistance of your skin. A skin conductivity sensor applies a positive and negative voltage to two parts of your body and measures the amount of current passing between those two points. The more current that gets through, the less resistance (more conductance) your body has, the more sweat you are assumed to have recently produced. The locations of the electrical contacts can vary, but a common choice is on two adjacent fingers. There isn't a whole lot of specialized equipment needed to make a skin conductivity sensor, so I decided to make one for my lie detector.

There are of course other bodily signals that can be monitored and loosely associated with the act of lying, but many require specialized equipment. For a DIY lie detector, these other methods require significantly more effort than a skin conductivity sensor requires, which is significantly more than I was willing to provide.

Construction

I decided to go with the two-finger approach to the skin conductivity sensor, where metal pads are attached to two adjacent fingers. I started by taping some wires to pieces of aluminum foil that could be wrapped around a finger.

Starting real simple here.

That was easy.

Metal conducts electricity. If you attach two metal pads to your fingers with the intention of measuring the resistance between the pads (your hand), you need to make sure the pads on each finger don't touch each other. The current that would be sent through your hand would instead prefer to travel directly between the two touching pads, causing the sensor to see an extremely high conductivity. This would throw off the lie detector and make it think your hand has suddenly turned into metal, which is probably a sign of lying. I needed a way of covering the metal pads to make sure they couldn't touch. I considered using medical tape to keep the pads on, but wanted a less permanent and hair-removing solution.

When I said that I was going to limit my building to only use parts I had lying around my apartment, I was telling the (polygraph approved) truth. What I didn't mention is that I have a 3D printer in my apartment and a few kilograms of printing plastic, so I wasn't really hurting for custom parts. I decided the best way to make sure the pads didn't touch was to print a custom finger-holder with the metal pads taped to the inside. I got to work on the design.

Definitely the hardest project I've done. Yep.

Perfect fit.

I attached the metal pads to the inside walls of each finger and started on the sensor electronics. Some basic resistance measurements using a multimeter tell me that I should expect the resistance between the two pads to be anywhere from 30 to 300 kΩ. Since I was able to vary the measured resistance by more than 50% just by moving the contacts around, I decided to forego any amplifiers in the sensing circuit and just use a voltage divider to get a 0 to 5V signal from the sensor that a simple Arduino controller could read. It may not be the most elegant or sophisticated sensor setup, but that's lie detectors for you.

Data Collection

With an Arduino reading the voltage from the sensor voltage divider, I recorded some data with the finger brace on and tried to manipulate the results with sheer willpower.



On the vertical axis of the previous plot, I've shown the resistance between the finger pads inferred from the voltage measured from the voltage divider. The arrows indicate instances when I attempted to elicit a skin conductivity response without significant movement. Ideally, I would have tested this by lying to an interrogator. But since I had no imposing figures around at the time, I attempted to trigger a response by thinking about stressful things like paying taxes or that dream I had where I showed up at middle school in just my underwear. The plot above shows that I was indeed able to trigger a response.

A stressful event seems to correspond to a decrease in skin resistance that persists long after the stressful event has passed. This is likely due to sweat that accumulates in small amounts underneath the sensor pads and cannot evaporate easily. To allow the lie detector to automatically determine whether or not a lie is being told, I put a condition on the first derivative of the skin. When the resistance dips down at a specified amount per second or more, the code decides I am lying.

Show/Hide Code

So I guess that's it. I set out to build a simple lie detector and I did it using some cheap parts and an Arduino controller. It took maybe 3 hours of work, plus an hour or so of printing time for the fancy finger brace. I've basically taken a project that has been done quite a few times already on the internet and added a dash of 3D printing to make it trendy. Is this really something that deserves a write-up? No, I can do better.


Good Cop, Bad Cop

TV and movies have taught me that anytime a suspect is interrogated, there needs to be two people asking questions. One plays the part of the bad cop who tries to break the suspect's resolve and get to the bottom of things as quickly as possible. The other is the good cop who offers the suspect salvation from the bad cop if they would only give up the truth. In order for my lie detector to be an effective mechanism for divining the truth, it needs a set of interrogators.

But before I dive headfirst into building a pair of human-sized bipedal robots with mustaches, I should try to boil down the idea of the two interrogators to their basic concepts. For example, boiling someone down into their constituent parts might be something the bad cop threatens to do to the suspect. They don't necessarily need to be strong with words, just strong with their hands. In fact, the only thing the bad cop really needs is a hand. To slap people with.

To build the bad cop, I needed a motor, a motor driver, an arm, and a hand. I happened to have a set of decently powerful motors and an accompanying motor driver from an older project of mine where I tried to motorize a large wooden cart. The project was a failure, but I see no need for the motors to live in shame forever. I adapted the old Arduino controller I had used for that project to simultaneously control a motor and read skin conductivity measurements.




The driver could supply enough voltage and current to the motor to make it swing around at a good slapping speed, but I needed a way of attaching an arm. While I've had good luck finding a few motor adapters online to attach wheels and such to motors, I have yet to see a motor adapter for a slapper-arm. So I designed and printed one myself. The arm would be made of two thin plastic rods I had sitting around, so the adapter needed to have two mounting holes for the arm and a set of screw holes for the motor drive shaft.



The strange V-shape is partially due to the limitations of 3D printing. Prints need to be fairly simple, strong, and preferably without any significant overhangs. The two plastic rods for the arm slide into the sets of three holes in the vertical parts, and screws are used to secure the bottom to an existing metal motor adapter.

The wonders of 3D printing.

I then glued the plastic rods in the adapter and glued a glove filled with cotton balls at the end of the rods. Ideally the glove should be filled with sand or rocks or something, but this was just a prototype. I would attach the motor to the wooden frame I made for my LED Planet project, since I'm all about reusing materials. Also, I didn't want to spend time or money on this.

All it needs now is an awesome mustache.

The programming for the bad cop was fairly simple. It would monitor the suspect, wait for a lie to be detected, then slap the suspect. After waiting a second or two, it would repeat this process forever. Testing the speed and intensity of the slapper was easy since it doubled as a high-fiving machine.



Thoughtful Questions

The next step was to create the good cop. This half of the interrogation team needed to offer questions to the suspect in a non-accusing way. It didn't need to sound threatening, as the threat came from a glove being swung around on a wooden frame. Not only did it need a calm, unwavering voice to question the suspect, it also needed to come up with appropriate questions to ask. To accomplish these tasks, I let my laptop play the part of the good cop. It had the processing capacity to form sentences and play them through the built-in speakers. You can't make a powerful investigative duo without subtle non-verbal language, so I used a USB-to-FTDI cable to allow the good cop and bad cop to communicate.

The text-to-speech was done using the free and open source tool eSpeak. In a unix terminal, turning a string of text immediately into audio that plays through your default audio channel is simple:

> espeak "Hello world."

There is an API for eSpeak, so commands like this can be translated into C code. Unfortunately, it's so much easier to just do a system() call and be done with it:

With that, my good cop code could talk in a calm and controlled voice to the suspect. To help placate the suspect, I added a simple visual to the code.

The final step in constructing the good cop program was to give it a method for generating appropriate questions. First, I hand-coded a couple softball questions that the good cop could start out with just to get a baseline reading for the polygraph. Then, I handed the script off to my girlfriend who wrote a few more lines that I couldn't read beforehand. If I knew when the harder questions were coming and what they were, I might have been able to skew the polygraph readings in my favor. I wanted a truly authentic experience, so surprise was important.

Suspects need alibis. The interrogators need to ask for an alibi. It seems common to ask the suspect about their actions on a particular date, so I added an automatic alibi-asker. It picks a random date within some range of years and forms a sentence that asks the suspect about the date.

I'm not entirely sure what instance the interrogation team would be referring to, but you can't go wrong with a random number generator asking the questions. Next was to add a few more human-generated questions that I didn't have to come up with. I needed a large repository of good, thoughtful questions, and the answer was obvious: Yahoo Answers. This is a site that allows users to pose questions to the public and rate submitted answers. It's full of all kinds of wonderful questions, some of which are collected in aggregate sites. I wrote another block of code to load a specified page of Yahoo Answers, parse through it for questions, then ask them sequentially.

Show/Hide Code

With all of the questions queued up, I positioned the good and bad cops for optimal interrogative power. I bravely volunteered myself for questioning by my own creation and recorded the results:




As you can clearly see, the full interrogation system worked perfectly. The skin conductivity sensor provided accurate measurements for the low-level polygraph hardware and it was able to trigger the slapping response at appropriate times.

I think this project has shown that an effective polygraph can be constructed using everyday materials like aluminum foil and 3D printers. I fully support any hobbyist following my lead and using their home-made polygraph to seek out the truth. Just remember, a useful lie detector relies on the understood and sometimes fully-realized threat of physical or emotional harm.