Building your DeepPiCar

---

I. Bill of Materials

• For this project you will need the following materials

Component	Link	Price (USD)
Male to Female Jumper	Link	$7
MicroSD Card (32 GB)	Link	$7
New Bright 1:24 Scale RC Car	Link	$20
Pololu DRV8835 Motor Driver	Link	$15
Pololu Power Cable	Link	$2
Power Bank	Link	$17
Raspberry Pi Zero 2 W	Link	$25
Raspberry Pi Zero Camera v1.3	Link	$7
USB Micro to USB	Link	$7
Total		$107

• Additionally, you will need to 3D print these CAD models to mount the hardware to the car.
  • Main hardware mount
  • Thread plate (NOTE: This thread plate is designed for m2.5 screws. Feel free to create other plates if you wish to use different screw types.)
  • Camera mount
  • Camera cover

II. Construction

Once you have gather all required materials, follow the build video below to assemble the car.

III. Software Setup

The DeepPicar repository is built and tested on the Raspberry Pi OS (Legacy) Bullseye. You can flash your Raspberry Pi Zero using the official Raspberry Pi Imager, available here.

Install DeepPicar Repo

Clone the DeepPicar repository and install dependencies:

git clone --recurse-submodules --depth 1 https://github.com/CSL-KU/DeepPicar-v3
cd DeepPicar-v3 
sudo apt update
sudo apt install libatlas-base-dev
sudo apt install libopenblas0
sudo apt-get install python3-opencv
sudo pip3 install -r requirements.txt

---

Configure Drivers

Edit the params.py file to select the correct camera and actuator drivers.
If you are using parts from the build list, you can skip this.

camera = "camera-webcam"
actuator = "actuator-drv8835"

---

Setup Driver for Pololu DRV8835

To install the Python driver for the Pololu DRV8835 motor controller:

cd drv8835-motor-driver-rpi
sudo python3 setup.py install

---

Setup Gamepad Support (Logitech F710)

If you’d like to use a gamepad for data collection, set up the inputs Python package:

cd inputs
sudo pip3 install .

IV. Test

Login to your Pi

SSH into your Pi over local WIFI

Before you start driving your car, make sure to enable legacy camera support.

$ sudo raspi-config

Navigate to interface options, select legacy camera, and click on yes.

Start the control script

$ cd DeepPicar-v3
$ sudo nice --20 python3 deeppicar.py -n 4 -f 30

Keyboard controls:
A: move forward
Z: move backward
S: stop
J: turn left
K: center
L: turn right
R: start/stop recording
D: turn on DNN

Use the keys to manually control the car. Once you become confident in controlling the car, collect the data to be used for training the DNN model.

The data collection can be enabled and stopped by pressing R. Once recording is enabled, the video feed and the corresponding control inputs are stored in out-video.avi and out-key.csv files, respectively. Later, we will use these files for training. It can be downloaded using scp commands.

Each recording attempt will overwrite the previous.

Compress all the recorded files into a single zip file, say Dataset.zip for Colab.

$ zip Dataset.zip out-*

Train the model

Open the colab notebook. Following the notebook, you will upload the dataset to the colab, train the model, and download the model back to your PC.

Open In Colab

After you are done training, you need to copy the trained tflite model file (large-200x66x3.tflite by default) to the Pi using scp commands.

Autonomous control

Copy the trained model to the DeepPicar.

$ cd DeepPicar-v3
$ sudo nice --20 python3 deeppicar.py -n 4 -f 30

Enable autonomous driving by pressing A to go forward then D to start the DNN.