Skip to content

gather data

Josh edited this page Jul 22, 2019 · 7 revisions

Gathering Training Data

One time setup

In order to keep the training data organized and in the same format so we call can benefit from your AWESOME driving skillz we have made some helper scripts. These require some one-time setup steps, but they should make life a bunch easier moving forward.

  1. Edit ssh_to_car.sh: This script adds the correct parameters to allow you to ssh to the car (Pi/Nano) but also allow the car to send data back via rsync. All you need to do is change the CAR_ADDRESS variable to [email protected]. ie: My username on the Pi is josh and my Pi's name on the network is dingo1 so I would set [email protected].

  2. Edit info.json: This file should be edited anytime you have changed the type of track or location you are gathering data in. It is used to keep track of information about the training data contained in a tub. ie: If I was training on a 40cm wide track on a timber floor in Buckingham Palace with thick green tape, no centerline my info.json would look like this:

{
  "count": -1,
  "date": "SET AUTOMTICALLY",
  "tape": "thick green",
  "floor" : "timber",
  "location": "Buckingham Palace",
  "centre line" : false,
  "lane width" : "40cm"
}

Now We Can Gather Data

  1. ssh to the Pi: To ssh to the pi we have a simple bash script so you don't need to remember the command. It's pretty simple but all the little things tend to add up.
cd dingocar
./ssh_to_pi
  1. cd play/roba1/mycar

  2. Fill in the info.json: The info.json is used to help keep track of when a dataset was gathers and what type of data is in the tub that is gathered. I recomend you fill this is at the start of a labelling session (or whenever your track or track environment changes). A sample info.json is below:

{
  "count"       : SET AUTOMATICALLY,
  "date"        : SET AUTOMATICALLY,
  "tape"        : "thick yellow",
  "floor"       : "timber",
  "location"    : "bill gates' house",
  "centre line" : true,
Clone this wiki locally