This is a collection of Jupyter Notebooks I use to train models for DIY Robocar Virtual Race League events. They are adapted from my GeneralAssembly Data Science Immersive capstone project.
My computer runs Ubuntu 20.04. You may have to change the directory commands to suit your operating system.
This should work with any Python version 3.6 or higher (due to the f-strings), but I haven't done any testing.
Input data configuration is an artifact of the way my client records things. 00_Data_Processing.ipynb expects a .CSV file with the following columns:
- steering_angle
- throttle
- speed
- image*
- hit
- time
- accel_x
- accel_y
- accel_z
- gyro_x
- gyro_y
- gyro_z
- gyro_w
- pitch
- yaw
- roll
- cte
- activeNode
- totalNodes
- pos_x
- pos_y
- pos_z
- vel_x
- vel_y
- vel_z
- lap*
With the exception of image and lap, these are simply the data contained in the telemetry messages sent by the Donkey Simulator.
* image: This is the filename of the image corresponding to each record. This is in place of the base-64 encoded string sent by the simulator. The images themselves are stored in another folder.
* lap: This is the lap from which the record was taken. The sim does not provide lap number in its telemetry messages; my client tracks them separately.
To plot trajectory, you need a dataframe with pos_x, pos_z, lap_time, and whatever you want to use as a hue for colormapping. By default, I use steering, throttle, speed, and activeNode.
- Please note that I have renamed the "steering_angle" column as "steering", as is only right and good.
