- Ari Cortes – ECE Senior
- Antoine Laget – CSE Senior (UCSD Extension)
- Kevin Lam – MAE Senior
- Jack Ringelberg – MAE Senior
In this class, students are tasked with programming a remote control (RC) car to navigate a track autonomously. This is first accomplished by using deep learning to train an AI model with the Donkey Car framework, and then tackled using Robot Operating System (ROS) to implement image processing and lane-following algorithms. In both cases, training and tuning result in a lot of eccentric behavior and crashes, so an emergency stop is implemented to minimize accidents. Currently, a relay controlled by a wireless clicker disables the PCA9685 pulse-width modulation (PWM) board, stopping steering and throttle commands from reaching the servo and motor. This assembly is not ideal because it is bulky, requires a lot of jumper wires, and causes the car to coast to a stop rather than brake. The goal of this project is to replace the relay and PWM board assembly with a single ESP32 wi-fi capable microcontroller. The ESP32 will generate PWM signals to control the servo and motor and will receive emergency stop commands through wi-fi from a user's phone or computer.
Figure 1: Old and New Emergency Stop System
- ESP32 generates PWM signals based on commands from Jetson Nano to control the servo and motor.
- ESP32 functions as an access point for a phone or computer to connect to.
- Website with a red button activates an emergency stop.
Nice to Haves
- Heartbeat and watchdog to shutdown car when wi-fi connection or serial connection with Jetson is lost.
- Expand website to include additional functions beyond emergency stop.
(QUICK VIDEO OF PROJECT (I (JACK) CAN DO))
Figure 2: Team 2 Car
The major components of the mechanical design include the baseplate, camera mount, and Jetson Nano case.
Starting from a high-contrast image of the car chassis, the baseplate was designed to conform to the existing body shape of the RC car. A central slot allows for easy wire passthrough for cameras and circuitry. A reversible design allows for ease of electrical debugging, and once that's working, simply flipping over the plate protects the electronics from collisions.
Figure 3: Baseplate Design
Multiple camera mount design iterations were tested over the course of the quarter. Starting with an adjustable design, once an ideal camera angle was chosen, a sturdy rigid mount was used to provide ample camera protection.
Jetson Nano Case
Based on a Thingiverse design, mounting holes were added for fixing the case to the baseplate.
Figure 5: Jetson Nano Case Modified from Thingiverse Design
The car's electrical assembly consists of eleven main components:
- Jetson Nano – The single board computer (SBC) in charge of controlling the remote control car.
- 3 Cell LiPo Battery and Alarm– The power source for the car. An alarm is attached to notify the user when the battery charge has depleted.
- Power Switch – Switches power to all components on the car except the electronic speed controller (ESC).
- DC-DC Converter – Regulates the battery voltage (anywhere from ~10-12V during operation) to a constant 5V.
- USB Camera – Camera connected to the Jetson via USB cable to provide a live video feed.
- Servo – Steers the car.
- Electronic Speed Controller (ESC) and Switch – Controls the DC motor based on commands from the PWM board and provides power to the servo. The ESC can be switched on or off.
- Brushless DC Motor – Drives the car's four wheels.
- PCA9685 Pusle-Width Modulation (PWM) Board – Receives commands from the Jetson Nano via I2C communication to control the steering and throttle. Generates PWM signals to send to the servo and to the ESC to be then sent to the motor. Also provides power for the status LED light.
- Status Light-Emitting Diode (LED) – LED light denoting enabled or disabled state of the car.
- Emergency Stop Relay – The relay is connected to a wireless remote which switches 3.3V power from the common (CO) terminal between the normally closed (NC) and normally open (NO) terminals. The disable pin (OE) of the PWM board is connected to the NC terminal, stopping generation of PWM signals when switched and therefore shutting down the steering and throttle. To denote the enabled and disabled conditions, the blue light is connected to the NC terminal of the relay and the red light is connected to the NO terminal.
Figure 6: Car Wiring Diagram with Original Hardware
The ESP32 (component 9* in the new diagram) replaces components 9, 10, and 11. The ESP32 communicates with the Jetson Nano via serial communication through a USB cable. Pin 0 and 4 of the ESP32, which are capable of generating PWM signals, are connected to the PWM inputs to the servo and ESC, respectively. The 6V power output from the ESC is connected to the servo. The servo, ESC, and ESP32 are all grounded to the Jetson.
Figure 7: Car Wiring Diagram after ESP32 Implementation
Git repository containing all source code can be found here.
ESP32 (LOLIN32) Code
Websockets and Syncronization
Figure 8: Website User Interface
PWM Signal Generation
For the servo PWM signal generated at pin 0, the command ledcWrite was used to generate a 3.3V PWM signal of varying duty cycle based on a normalized steering command between -1 and 1 from the Jetson . For a PWM frequency of 300 Hz, a 33% duty cycle was found to correspond to full left steering and 58% corresponded to full right. Visualization of this signal is provided below:
Figure 9: Servo Left and Right Steering Signal
It was found that the ESC does not function with a fixed-frequency PWM signal. Instead, the ESC requires a 20 millisecond low period followed by a 1-2 millisecond high (3.3V) pulse. 1 millisecond high corresponds to full reverse rotation of the motor and 2 milliseconds corresponds to full forward rotation. To generate such a signal, the writeMicroseconds command from the ESP32Servo Library was used to modulate the duration of the 3.3V pulse generated at pin 4. The visualization of this signal is shown below:
Figure 10: Motor Full Reverse and Forward Signal
Jetson Nano Code
On the Jetson Nano, provided image processing and vehicle control nodes output throttle and steering topics. These topics are parsed by our ESPcomms_ROS node and formatted as a JSON string for serial passing to the ESP32.
Figure 11: rqt_graph of ROS nodes on Jetson Nano
Donkey Car Deep Learning Autonomous Laps
ROS Autonomous Laps
ESP32 with E-Stop
Advice and Suggestions for the Future
- Build to Crash - No matter how carefully you drive, you will probably crash a lot! A lot of collisions occur between cars and are often not your fault. The best way to prepare for this is to ensure your mechanical parts are beefy and electrical components are covered.
- Get Driving ASAP - Training a deep learning model at the tent track was a particularly difficult task due to the constantly changing lighting conditions. It was critical to start training early to ensure there was enough time to work out any kinks and develop a robust model. Listen to Professor Silberman and send your parts out to be manufactured in the first week!
Future Suggestions for ESP32 Use
- Expanding ESP32 Website Functionality - Possible functions to add to the website include live calibration of steering and throttle, live color filter adjustment, and live PID controller tuning.
- Creating Class for DonkeyCar Implementation -
Acknowledgements and References
- Team 1 - Thank you for helping us determine the proper PWM signals to send to the motor and servo.
- Dominic and Haoru - Thank you for debugging with us throughout the quarter.
- Professor Silberman and Professor de Oliveira - Thank you for providing a priceless learning opportunity!