Obstacle Detection System

Motivation

Self-driving car usually reacts faster than human when encountering unexpected event. However, self-driving car is not perfect, faults and blind spots still might occur. Furthermore, for some road situations such as blind corner, it will be risky to solely rely on self-driving system. Road-side devices can provide useful information to both human-driving cars and self-driving cars. As an essential part of smart city, road-side devices that can detect and notify road events and conditions should be prevelant in the near future. Therefore, I decided to build an simple obstacle detection system in this project; its basic concepts might be useful in the future.

Methods

This project is composed of a development board, a self-driving car, and an ultrasonic ranging module. In real world, it is too costly to put development boards on road-side, and the sensor might not be suitable for real roads. Self-driving car and development board are both connected to same Wi-Fi network. The propose of this project is to test out different combinations and validate ideas only.

Hardware

The overall hardware arrangement is shown as the image below.

• STMicroelectronics STM32L4 Discovery Kit IoT Node (B-L475E-IOT01A)¹
  • Laser ToF and gesture detection system (VL53L0X, onboard)
  • Ultrasonic ranging module (HC-SR04)
• PiRacer AI Kit
  • Raspberry Pi 4 Model B
  • Canvas circuit
  • The car can drive around the circuit automatically

Software

• STM32

  • Send socket when car passed checkpoint.
  • Socket contains obstacle and car information.
  • Assume car passed obstacle successfully after it goes into obstacle warning area for 5 or 10 seconds.

• Raspberry Pi

  • isCar & isObstacle: Manual mode
  • isCar & ~isObstacle: Auto mode
  • ~isCar: Auto mode

Results

The obstacle detection system has been built up successfully. Since I am still unfamiliar with donkey car, this system cannot change driving mode or throttle settings while the car is running; it can only stop current driving process and start a new driving process which has different settings.

Usage

STM32 Settings

1. For proper EventQueue handling, go to mbed-os/platform/mbed-lib.json and change the value of callback-nontrivial to false.

2. (Optional) To make console print out floating point numbers, go to mbed-os/targets/targets.json and change the value of printf_lib to std.

3. Edit mbed_app.json to include the correct SSID and password.

4. Set SocketAddress addr("IP",port) in main.cpp.

Raspberry Pi Settings

1. Setup Wi-Fi connection. The username of Raspberry Pi is pi, and the password is eslab305.

2. cd mycar.

3. Set HOST and PORT in server.py.

4. Put all files in RPi directory to this directory (mycar).

Run Program

• Testing Mode
  1. On Raspberry Pi, run server_test.py using command python server_test.py.
  2. Reset STM32.
  3. Raspberry Pi's console will show expected driving mode under different car-obstacle situations.
• Auto Driving Mode
  1. On Raspberry Pi, run server.py using command python server.py.
  2. Reset STM32.
  3. The car will stop and start new driving process if it should.

Demo Videos

Please refer to server_test.mp4 and server.mp4.

Discussion

• Ultrasonic ranging module is quite stable.

• Laser ranging sensor needs tuning

• The workaround of being unable to control drive mode directly had no success.

• For STM32, all functions are in main.cpp;

Future Work

• Try other sensing method; the scanning angle of current method is small.

• Determine whether the car passed obstacle detection area or not, instead of just waiting for 5 or 10 seconds.

• Allow multiple cars to be on road.

• Active obstacle information receiving mode: car request obstacle information actively when it goes into obstacle warning area.

References

1. PiRacer AI Kit 组装教程

2. GitHub: 2020-GRD-cultivator

3. How to terminate a subprocess in Python

Footnotes

1. In this document, STM32 refers to this device. ↩