This is a Python Rover project which currently uses 3 ultrasonic sensors and a magnetometer to navigate. Its a fairly specific thing I used as a learning tool perhaps someone may find as a useful example.
It is capable of streaming basic OpenCV image processing done onboard at roughly 700ms frequency (very slow), helper scripts are included to calibrate a lens for image processing.
It does not navigate by the houghline detection it performs due to the capability being limited by the visual frequency. Of roughly 2.5fps which would be impacted further by additional processing.
I am currently developing a rover on a far more capable vehicle platform and compute platform.
The lens calibration seen in this stream is atrocious. This was a 30s job. ROI cropping failed accordingly. For best results move rover and hold still for a second in each position. You will get a proper square feed. For perfect results, symmetrically align all your pictures along a predefined schema and take single stable shots. (need to refactor calibration_capture to wait for keypress then capture)
Raspbian Stable/Raspberry Pi OS Bookworm
Raspberry Pi OS with desktop
Release date: March 15th 2024
System: 64-bit
Kernel version: 6.6
Debian version: 12 (bookworm)
Size: 1,105MB
Install everything in Linuxfile with apt
for package in `cat Linuxfile`; do
apt -y install $package
done
Create a venv for the files, The drivers used require root access.
cd ~
git clone https://github.com/russmac/python_rover robot
cd robot && python3 -m venv venv
. venv/bin/activate
pip3 install -r requirements.txtEdit instance/config.py
camera_selection = "picamera_v2_vga"
camera_configurations = {
"picamera_v2_vga": {
"id": "picamera_v2_vga",
"fisheye": True,
"canny_auto_levels": True,
"stream_canny": False,
"camera_x": 640,
"camera_y": 480,
"canny_params": {"upper": 160,
"lower": 40},
"hough_params": {"minLineLength": 100,
"maxLineGap": 5,
"probabilistic": True},
},
}
motor_configuration = {
"enable_pin": 17,
"slew_pin": 27,
"left": {
"pwm_pin": 13,
"forward_pin": 6,
"reverse_pin": 5,
"state_flag_pin": 11,
},
"right": {
"pwm_pin": 12,
"forward_pin": 20,
"reverse_pin": 16,
"state_flag_pin": 26,
},
}
# Tweaking these values can dramatically effect navigation.
My specific rover can competently navigate around chair legs and furniture under and rarely if ever gets stuck.
ultrasonic_configuration = {
"ussl_pin": 7,
"ussc_pin": 25,
"ussr_pin": 8,
# How many cm is considered clear from center sensor
"clear_distance_c": 35,
# How many cm is considered clear from side sensors
"clear_distance_side": 35,
# What cm is a clear path.
"path_distance": 48,
}
And to start up
python3 start_threads.pyThe app looks for a camera calibration in ./config matching the camera id in instance/config.py , The same config is used in the calibration helper scripts so the files should be correctly named.
The robot will concurrently look for a locally saved Bosch BNO055 calibration, Whichever parts are present will be used, You will be prompted to calibrate the rest.
https://www.youtube.com/watch?v=Bw0WuAyGsnY
This can "glitch" in that the sensor rejects the saved calibration due to some validation. At time calibration can seem like magic and some commenters noted dancing with the sensor (robot) is more effective. I have not tried this.
root@media-desktop:/robot/python_rover_robot# python3 threaded.py
WARNING | Orientation | Expecting value: line 1 column 1 (char 0)
INFO | Orientation | Calibrate now!
INFO | Orientation |
Gyroscope: 0
Acceleromter: 0
Magnetometer: 0
...
Follow calibration video instructions
...
INFO | Orientation |
Gyroscope: 3
Acceleromter: 1
Magnetometer: 3
INFO | Orientation | Fully calibrated
INFO | Orientation | Switching to config mode
INFO | Orientation | Saving to cache: GYR_OFFSET_X_LSB
INFO | Orientation | 0
INFO | Orientation | Saving to cache: GYR_OFFSET_X_MSB
INFO | Orientation | 0
INFO | Orientation | Saving to cache: GYR_OFFSET_Y_LSB
INFO | Orientation | 3
INFO | Orientation | Saving to cache: GYR_OFFSET_Y_MSB
INFO | Orientation | 0
INFO | Orientation | Saving to cache: GYR_OFFSET_Z_LSB
INFO | Orientation | 0
INFO | Orientation | Saving to cache: GYR_OFFSET_Z_MSB
INFO | Orientation | 0
INFO | Orientation | Saving to cache: MAG_OFFSET_X_LSB
INFO | Orientation | 24
INFO | Orientation | Saving to cache: MAG_OFFSET_X_MSB
INFO | Orientation | 0
INFO | Orientation | Saving to cache: MAG_OFFSET_Y_LSB
INFO | Orientation | 86
INFO | Orientation | Saving to cache: MAG_OFFSET_Y_MSB
INFO | Orientation | 0
INFO | Orientation | Saving to cache: MAG_OFFSET_Z_LSB
INFO | Orientation | 215
INFO | Orientation | Saving to cache: MAG_OFFSET_Z_MSB
INFO | Orientation | 0
INFO | Orientation | Saving to cache: ACC_OFFSET_X_LSB
INFO | Orientation | 200
INFO | Orientation | Saving to cache: ACC_OFFSET_X_MSB
INFO | Orientation | 255
INFO | Orientation | Saving to cache: ACC_OFFSET_Y_LSB
INFO | Orientation | 22
INFO | Orientation | Saving to cache: ACC_OFFSET_Y_MSB
INFO | Orientation | 0
INFO | Orientation | Saving to cache: ACC_OFFSET_Z_LSB
INFO | Orientation | 227
INFO | Orientation | Saving to cache: ACC_OFFSET_Z_MSB
INFO | Orientation | 255
INFO | Orientation | Saving to cache: ACC_RADIUS_LSB
INFO | Orientation | 232
INFO | Orientation | Saving to cache: ACC_RADIUS_MSB
INFO | Orientation | 3
INFO | Orientation | Saving to cache: MAG_RADIUS_LSB
INFO | Orientation | 189
INFO | Orientation | Saving to cache: MAG_RADIUS_MSB
INFO | Orientation | 2
INFO | Orientation | Configured calibration cache
INFO | Orientation | Switching back to NDOF mode
INFO | Driver | Sensor data populated
WARNING | Driver | I think its clear... driving
https://docs.opencv.org/2.4/doc/tutorials/calib3d/camera_calibration/camera_calibration.html#
- Run ./calibration_capture.py on your Rpi and take your captures, This process is really important to get right as without an undistorted image machine vision algorithms are ineffective.
- You need to rsync or scp ~/robot/calibration to ~/samples/calibration on your local machine when you're done.
- You then run calibrate.py, it will show you progress as it attempts to find chessboard corners.
- Successfully chess-boarded calibration images will be saved (and drawn on) under samples/results , Failures will be deleted from samples/calibration.
- Numpy binary arrays will be saved under samples/config , You need to SCP or Rsync these to ~/robot/config (including sample_image.jpg).
Navigate to http://robot:8000 where robot is your robots IP address.
- Raspberry Pi 4B (used in my project)
Other options;
- Raspberry Pi 3B
- Nvidia Jetson Developer kit
- Nvidia Xavier NX
- DFRobot "Devastator" platform (any two motor rover will do)
- Polulu Dual MC33926 Motor Driver
- 3x DFRobot URM37 v5.0 Ultrasonic rangefinders. (these are fairly inaccurate but do the job)
- Adafruit ADS 1015 12 bit ADC
- DFRobot 10DOF AHRS (Bosch BNO055 & BMP280)
- PiCamera V2
- Adafruit PowerBoost 1000 Charger 5v 1A
- Polymer Lithium Ion Battery (LiPo) 3.7V 6000mAh
- 3A USB plug pack for charging / power supply while operating.