Corral

Offline training suite for autonomous RC vehicles

• Corral, offline vehicle training suite

DataSet class

Enforces contract for further offline processing pipeline

Strictly aggregates the numpy images and the saved records and signals on the timestamped data

Transforms underlying records

Flexible masking (cleaning) of data set by;

DataSet.mask_last_Ns()

DataSet.mask_null_throttle()

Masked 459 timesteps throttle<0.1, current cover: 7.8%
Masked 6181 timesteps from 2018-09-12 16:23:26.186000 to 2018-09-12 16

• Start - stop keyframe markers (datetime stamps)
• State values i.e. throttle < 0.1

Plotter class

• Operates on DataSet to plot summary histograms, charts etc.
• Operates on DataSet to generate analysis frames with a HUD overlay

DataGenerator class

Operates on DataSet to serve batches to Keras

Trainer class

Operatates on DataSet to train a Keras model

Saliency class

Operates on DataSet to visualize the network activations

VideoWriter class

Uses ffmpeg to write a video from a directory of frames