Skip to content

The repository contains the implementation of transfer learning, continuous and discrete CCNs for lane navigation of a SunFounder PiCar.

Notifications You must be signed in to change notification settings

akinolawilson/PiCar-Autonomous-Lane-Navigation

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

12 Commits
 
 
 
 
 
 
 
 

Repository files navigation

PiCar: Autonomous Lane Navigation


Overview


This repository contains the notebook that sees the implementation and optimization of a convolutional network. The underlying task is to prediction two quantities; steering angle and speed of a SunFounder PiCar.

Data collection
The data collection was performed in segments, where each particular scenario of the final test enirvonment was targeted. The remote controlled car was manuvoured via external controls around the following tracks:



Where the following situations were considered:

  • Clear ring
  • Clear T-junction
  • Ring track with obsteciules
  • T-junction with traffic lights
  • T-junction without stationarity

    In total, there are around 3000 training examples. The set is split using 70:20:10 ratio between the training, validation and evaluation sets. To investigate the raw datasets, please find them via this link.

File content


Data preparation

The raw files are separated via their original collection. There are around 13 folders. The data_exploration.ipynb notebook contains the sortation and exploration of the folders content, preparing it for the training process and visualizing descriptive statistics.

Modelling

The CNN_model.ipynb contains the fitting and optmization process of the CNN model. Tensorflows' hyperparameter API alongside the TensorBoard API has been used for the model optimization process.


About

The repository contains the implementation of transfer learning, continuous and discrete CCNs for lane navigation of a SunFounder PiCar.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published