Difference between revisions of "2021SpringTeam6"
|Line 37:||Line 37:|
=== Algorithm ===
=== Algorithm ===
== Demos ==
== Demos ==
Revision as of 22:13, 10 June 2021
One of the things that we noticed with the Donkey car and ROS car, that would be a major problem for an actual passenger vehicle is the lack of object avoidance or the ability to stop during emergencies. While more expensive sensors exist for implementing object avoidance, like LIDAR, we thought that less expensive solutions could be made. For implementing object avoidance on our car, we chose to go with the inexpensive HC-SRC04 Ultrasonic Sensor.
Project Team Members
- Dylan Perlson (ECE)
- Harrison Lew (MAE)
- Joshua Guy (MAE)
- Arman Mansourian (CSE)
NVIDIA Jetson Nano, Camera (get specific), Adafruit PWM Controller, Electronic Speed Controller, 12V-5V DC-DC voltage converter, Throttle DC motor, Steering Servo Motor, 11.1V 3S LiPo Battery, Voltage Sensor, Remote relay, HC-SRC04 Ultrasonic Sensor
We used the wiring schematic from Fall 2020 Team 2 to get a head start on our wiring and allow us time to implement our additional components. Our final circuit was similar to theirs except included the ultrasonic sensor that we implemented in our design with a voltage divider for conditioning the signal to a readable level for the Jetson Nano (insert diagram here)
For OpenCV, we had to modify several parameters for our implementation. We had to calibrate the pwm values for the throttle motor and for the steering servo and find appropriate control values for achieving the turning performance that we wanted. We also had to calibrate the color filters and max/min width of line detection. A consistent issue that occurred and had to be dealt with was the changing light conditions of the tent in which testing was done, meaning that the color calibration had to constantly change.
ROS (Robot Operating System)
For ROS, we based our implementation off of the code that Dominic provided. The code that was provided had nodes for throttle, steering, lane detection, lane guidance, and camera server. Combined, these nodes allowed the car to drive within the track and follow the lanes after proper color calibration. Our project expanded on this autonomous behavior by adding an additional node that used the ultrasonic sensor and relayed distance information for objects in front of the vehicle. The ultrasonic sensor node published distance readings and the lane guidance node subscribed to it.
The goal of our algorithm was to adjust the throttle so that the car would be able to slow down when an object approached and have an emergency stop for anything that abruptly comes in front. To do this, we used the Ultrasonic Sensor to continuously read the distance of any object in front of the car. Any object 150cm away or further did not affect the car. Any object detected between 150cm and 75 cm was linearly scaled down to a throttle of zero using a multiplier: alpha = (distance_detected - 75cm) / (150cm - 75cm). This way the car would come to a gradual stop if an object was detected far enough away. However, if an object fell in front of the car at 75cm or less, the car would immediately stop.
One problem we ran into was that the Ultrasonic sensor was very noisy and its values would jump around alot. To solve this, we added a moving average to the distance input to smooth out the signal. After adding this, the car would no longer accidentally come to an abrupt stop if there was any random noise that caused the signal to be lower than it should have been.