Servo controller

picar

Car with autonomous and remote-controlled capabilities controlled by Raspberry Pi 5 and a CH32V003 microcontroller. Features a custom made mainboard PCB responsible for controlling motors, servos and delivering power to the RPi 5.

It is capable of real-time remote control with a gamepad (e.g. XBox One gamepad) alongside a live WebRTC stream from its camera. Moreover, simple autonomous driving is possible by following an RGB tracking marker.

Version 3 Early (still the old frame from V2)

PiCar with the tracking marker next to it

Future ideas

• Add a ToF sensor and use it in tandem with the DepthAnything V2 depth estimation model to acquire real (approximated) depth information from the camera.

Hardware

• Battery pack composed of 3x 18650 cells connected in series. It has both XT60 connector for the main power and 4-pin JST-XH used for balancing.
• SG92R micro servo used for steering
• Raspberry Pi 5 8GB
• Raspberry Pi Camera HD v2 8MPx
• Custom PCB mainboard responsible for controlling motors, servos and delivering power to the RPi 5. (More on that below)
• Custom 3D-printed frame and steering system (files available in the "designs" directory)
• 4x 60x8mm black Pololu 1420 wheels
• 2x 6x13x5mm bearings for the front wheels
• 2x N20-BT07 micro 100:1 320RPM 12V motors
• M3 screws used for assembling the frame and wheels together

PCB mainboard

/pcb_main_board/pcb_main_board_schematic.svg

/images/pcb_main_board1.png

On-board microcontroller

CH32V003 MCU is the main controller of the whole board. It is responsible for controlling the motor drivers, servo motors and monitoring the battery voltage. It communicates with the Raspberry Pi 5 over I2C.

Battery to RPi 5 voltage step-down

The battery input voltage is first filtered by the input low-ESR capacitor and capped to 15V by a SMBJ15A TVS diode. Then the MP2338 IC is used to reliably step-down the battery voltage from 9V-12.6V range to ~5.1V that is then connected via a USB-C cable to the RPi 5. The MP2338's GND pin is connected to wide copper traces with vias in order to improve cooling.

Motor control

TB6612FNG IC is used to drive both of the main motors. The IC logic is driven by PWM signals generated by the on-board microcontroller. The PWM frequency can be adjusted from the software layer if the user wants to prioritize peak power or to minimize the audible switching noise. The lower the frequency, the more efficient the motors are, but the louder they become.

Servo control

The mainboard is capable of running 4 separate hardware PWM signals at once which are generated by the on-board microcontroller. The servo motors are powered with 4.8V generated by another MP2238 IC configured in the same way as the other one used for powering RPi 5, except for the output voltage. The servo section also has a 3-pin header for each signal channel in order to simplify the connections. There are more servo channels than are currently needed for future-proofing reasons. e.g. allowing the camera to have adjustable yaw and pitch in the future.

Wheels

The front wheels are fastened to their mounts with short M3 screws. Those mounts then clip into the ball bearings, working similarly to LEGO Technic connections. A pin with the correct diameter and side cuts is pushed into the bearing hole and holds firmly enough for normal operation. The rear wheels are press-fitted onto the motor shafts:

Front wheel assembly disassembled in Fusion 360:

Software and control

Frontend

PiCar can be controlled through a simple Websockets-based web interface (/car/software/index.html). When you run the Python "Remote Control" main loop an HTTP server starts up and reveals the visual interface at http://{raspberry_pi_5_ip}:8000. There are two sliders that control it but the better (and way cooler) option is to use a gamepad and control the car this way. The website automatically detects connected gamepads and lets the user steer by using both triggers for accelerating and braking and the left thumbstick for steering. It is also possible to set the motors PWM switching frequency.

There also is a viewport for watching the live feed coming from the camera connected to the RPi 5. The stream source must be a WebRTC source which can be created easily by using MediaMTX. You can find the working mediamtx.yml file I use to enable the stream at /car/mediamtx.yml. Once mediamtx starts with my provided config, the camera can be accessed by connecting via WebRTC to {raspberry_pi_5_ip}:8889/rpi.

Vision

Autonomous driving, for now, is a simple computer vision algorithm that searches for a special marker with three key colors: red, green, and blue on the live feed streamed from the Raspberry Pi Camera HD v2 8MPx.

The camera feed is first filtered bilaterally to remove small details and noise in the subsequent marker search. Then gamma is adjusted to increase contrast, and the resulting image is binarized so that the SimpleBlobDetector from the OpenCV library can detect three circles. Finally, the algorithm checks whether the circles are "inside" the potentially found marker—i.e., whether they are not shifted from the averaged center by more than one standard deviation. If they are not, we can be certain that the marker has been found.

Below is a visualization of how the marker is seen by the algorithm:

When the marker is found, the vehicle drives forward and turns so that the center of the marker aligns with the center of the camera's view. This is sufficient to effectively track the marker.

Future Vision Enhancement

I plan to use a ToF distance sensor in tandem with the computer vision model DepthAnythingV2, which estimates depth using a single camera. This combination will allow me to acquire approximate (yet sufficiently accurate) distance information of everything visible in the front camera. This data will enable shape detection by contours and autonomous obstacle avoidance.

Firmware

The CH32V003 microcontroller runs a firmware that acts as a PWM and motor control interface for the Raspberry Pi 5. The firmware exposes a register-based interface over I2C using slave address 0x17, allowing the RPi to elegantly control all aspects of motor and servo operation.

I2C Communication Protocol

The firmware operates as an I2C slave that receives write-only commands structured as [register_id, data_low, data_high]. Each register controls a specific aspect of the system:

Building the code

RPi 5 Python

Configure the virtual environment if you need and run pip install -r requirements.txt while being in the /car/software directory to download the necessary packages. Downlaod opencv as a standalone package. Once everything is downloaded, run python main.py and it should work fine.

Enabling the necessary RPi 5 options

To make I2C communication work properly, enable it either with the raspi-config command or by manually adding the necessary lines to the /boot/firmware/config.txt file:

dtparam=i2c_arm=on
dtparam=i2c_arm_baudrate=80000 # or any other as you wish
dtparam=i2s=on

Also enable the camera detection in the same way. Here is the line for /boot/firmware/config.txt:

camera_auto_detect=1

CH32V003 C Firmware

Downloading and building it through MounRiver Studio IDE is the easiest route to go. This repo already has project files that work out of the box when you open them with the IDE. Just press F7 and it should build perfectly fine.

Flashing it

You'll need a WCH LinkE programmer in order to flash the firmware to the chip. Once you have it, connect GND to GND and the SWDIO pin to the chip's SWIO pin according to its datasheet (It's pin PD1 for CH32V003A4M6). The chip must be powered either from the programmer's voltage pin or externally during flashing.

Now go to MounRiver Studio IDE and press F7 to build your project and then F8 to flash it having the WCH LinkE programmer plugged into your computer at the same time, note that flashing your code does not rebuild it automatically, you have to remember to press F7 each time.. You can also change the flashing configuration but it's almost always good to go by default. (called Download Configuration):

Do not press or hold the reset button during flashing (the NRST pin)

Archive

Take a look at the PROGRESS.md markdown file to see how the project evolved over time.