In this video, I'll be building
a robot using minimal effort by utilizing the conversational AI bot, ChatGPT.
I'll give it simple prompts and follow its instructions to try creating an autonomous,
line-following robot without doing any software programming, electronic circuit design, or
even CAD modeling. This video is sponsored by PCBWay.
Before we get too carried away blindly building robots for an artificial intelligence, I want
to start with something a little simpler. Let's see if ChatGPT can teach me how to
blink an LED on and off using an Arduino. I'll start off by asking what components I need.
Not only does it provide information I explicitly ask for, but it also inferred
that I may want instructions. Okay, let's put these to the test.
Step 1. Connect the positive leg of the LED, the longer
leg, to a digital output pin on the Arduino. Step 2.
Connect the negative leg of the LED, the shorter leg, to a resistor.
Step 3. Connect the other end of the
resistor to ground on the Arduino. Step 4.
Use the Arduino software to write a sketch that will turn
the LED on and off at the desired frequency. Wait, but how do I actually program an Arduino?
I have everything wired up. Can you write the program for me?
Brilliant. Not only does it provide code, it also
provides details on the uploading procedure. Let's give it a shot.
Copy the sketch into the Arduino IDE and click upload.
It works! Okay, let's take off the training wheels. Do you think you could tell me the steps to
building a line following robot using Arduino? Wow, this is amazing!
It's literally giving every step that I need. This honestly looks pretty robust. Let me go grab the components
that are listed here. Okay, so I have most of the components here
with me, but I want to see if I can substitute in the L298N motor driver instead of
using the recommended L293D motor driver. I also want to see if it will allow
me to use a set of AA batteries as opposed to a 9-volt battery.
Let's see what it says. At first pass, this looks like a brilliant answer.
It signs off on my requested substitutes, but this is also where ChatGPT slips up for the first time.
These statements about needing an external voltage regulator like a 7805 are actually false.
Since the L298N motor driver already has a 5-volt regulator on board.
In fact the 5-volt regulator on the board is literally the 7805
it recommended we add externally. It's asking for the exact component
that's already on the motor driver. I would say that this is the
biggest downfall of ChatGPT. When it's wrong about something,
it is confidently wrong. Let's see what happens when
I challenge that response. Thanks for the info, although I think the L298N
already has a 5-volt regulator on the board. Okay, this looks better.
I'll wire it up following the chatbot's recommendations.
Positive battery terminal to input VCC, negative battery terminal to ground, and then the 5-volt
pin from the L298N to the V-in pin on the Arduino. Wait a minute, this is another mistake.
The V-in pin on the Arduino that we're told to connect 5-volts to only has a recommended
range of 7-volts to 12-volts. Instead we really want to
connect to 5-volts directly. So this is another place that ChatGPT has
spoken incorrectly with confidence, but for the moment I'm going to just brush it aside.
Let's go ahead and finally connect ground to ground.
We now need to connect the motors. Again, the bot gives information that could
work, but definitely isn't the proper procedure. This method would only allow the motor to spin
in one direction, instead of both directions. This time I'm not going to brush it under the rug.
No ChatGPT, that is wrong!! I have to admit it's pretty neat how you
can have a conversation with the bot similar to a human interaction, and if you would
disagree on something you can rationalize your argument by presenting facts in a logical manner.
However this only works if you have pre-existing knowledge in the field.
To use ChatGPT effectively, you'll need to act as a curator, researching and validating
information from third-party sources. I'm going to skim over the next few questions
for the sake of brevity, but if you want to go through the entire chat log check
out the link in the description below. Basically I'm just asking it about wiring a
second motor and which pins on the Arduino I should connect to.
I then go on to ask about connecting the infrared sensor array.
Here's the final wiring of the robot. It looks good, but it's lacking a bit of form.
ChatGPT mentioned earlier in our conversation that I need a robot chassis.
Maybe it can help design and build one? My local makerspace has a laser cutter, so
let's design an SVG file for laser cutting. I'll start with a simple test by asking
for an SVG file of a 25mm square. ChatGPT gives me a set of instructions
to create a file for myself in Inkscape. While these steps do sound good, I'm lazy.
I want you to do it for me, ChatGPT. Can you write the SVG file for me?
Not only does it write the file, it remembers the laser cutter
context from the conversation and tells me to save it with a
.SVG extension in a notepad editor. And then it tells me that I can import
the file into the laser cutter software. Absolutely brilliant.
But let's now do something more useful. I want to create footprint files
for each of these components. The motor drive is a 43mm
square with 3mm mounting holes. Here I'm giving the chatbot the dimensions
and requesting it return an SVG file. Opening it up, and sure enough,
it matches the motor driver. I'm not going to bore you to death. This part that I'm skipping over quickly is
me just asking it to create footprints for the rest of the components, and
these are the resulting SVG files. Now with all the individual components drawn,
let's create a new file for the chassis itself, measuring 220 by 150mm
and import the footprint files. Another annoying limitation of ChatGPT is that
it truncates answers beyond a certain length. Fortunately I can ask it to just continue from the last line. Copying this answer again and opening it
up, I guess you could say it kind of worked. It imported all the footprints and ensured there
were no overlaps between any of the components. However, their placement was an exactly
conducive to efficient robot locomotion. I'm sure I could go back and forth with a bot
and move everything into place, but I ultimately decided it was more efficient just for
me to manually edit the files myself. And here's the final design after
everything was shuffled around. Throwing the file on the laser cutter,
This is what came out. Now this isn't actually the first time
I've built a line following robot. I actually teach an Intro to Robotics class
here in Vancouver, BC using this exact robot. And as a side tangent, if
you are based in Vancouver and would like to learn Intro to Robotics
from a real person and not an AI chatbot, then consider checking out some of my classes
that include robot kits like this one, link in bio.
But based on this experience, I also know that I need some 3D printed brackets to mount
the wheel and infrared sensor on. I initially tried re-creating these components
as STL files using chatGPT, but honestly, it was a hair pulling frustrating activity.
I couldn't even build a simple cube with closed faces.
I'm sure you could wrestle the chatbot into submission and get primitive STL files,
but I ultimately decided to cheat and just use the existing mounting brackets that I already had.
Please forgive me. This still counts as a mostly
AI-generated robot, right? And with this, everything was assembled together.
We're ready to program this robot and bring it to life using code generated
by an artificial intelligence. But before getting to that, we have a
segment from this video's sponsor, PCBWay. But this time, I'm going to
get ChatGPT to help me out. Write an ad read for a 30-second sponsored
segment from PCBWay in a YouTube video. Bring your ideas to life with PCBWay.
PCBWay is the go-to destination for fast turnaround times, affordable
prices, and high-quality PCBs. Whether you're a maker, hobbyist, or engineer,
their PCBs are designed to meet your needs. With a strong reputation for providing excellent
customer service, PCBWay is the perfect choice. So why wait?
Visit PCBWay today and get started on your next big project.
It's absolutely insane what this chatbot can do. I will quickly add that if you've never built
a PCB before and want to learn how to do that, I have an entire series on my
channel, so check that out if you want. But for now, let's move on to the software, and
this is the part that I was most worried about. But my fears were misplaced,
because check this out. Based on the electronics and the pin allocations
we discussed earlier, please write an Arduino program with five functions.
Drive forward, reverse, left, right, and stop. Include a variable for
storing the PWM drive speed. This program should loop through the five
different drive functions with a two-second delay between each.
In just several seconds, the chatbot has given me a program to test out the robot's
drive train functionality. As requested, it has five functions for each
of the directions and has a variable called DRIVE_SPEED for the PWM speed control.
Testing it out, and it's perfect. Okay, let's go for it.
The full robot program. Let's get it to add in the infrared sensor
input and write an algorithm to follow a black line over a white surface. We want to use the sensors to detect when
the robot drives off course by crossing over the line and then correct
steering to get back on the line. Let's give these instructions to
ChatGPT and see what it generates. Okay, so we have the full program, and I
see we have some conditional logic functions in here.
But will it work? Perfect.
It works so well, I'm so happy. The robot detects the line and then
turns to stay right on top of it. Honestly, I don't even know why Tesla is having
such a hard time creating full self-driving mode.
I should just give ChatGPT the code base. So I do concede that this robot
wasn't technically built by AI. I had to definitely lead a helping hand
and redirect it when it was left astray. But it was still wildly impressive
what we were able to achieve. Stay tuned for more projects,
I'll see you in the next one.
