https://github.com/adrianmarino/self-driving-car-model
Steering wheel angle predictor
https://github.com/adrianmarino/self-driving-car-model
autonomous cnn conda keras machine-learning opencv python self-driving-cars
Last synced: 2 months ago
JSON representation
Steering wheel angle predictor
- Host: GitHub
- URL: https://github.com/adrianmarino/self-driving-car-model
- Owner: adrianmarino
- Created: 2018-11-29T00:45:58.000Z (over 7 years ago)
- Default Branch: master
- Last Pushed: 2019-03-10T03:16:18.000Z (over 7 years ago)
- Last Synced: 2025-10-10T01:03:02.711Z (9 months ago)
- Topics: autonomous, cnn, conda, keras, machine-learning, opencv, python, self-driving-cars
- Language: Jupyter Notebook
- Homepage:
- Size: 129 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Self driving car model
* The model learn to control steering wheel angle.
* Throttle & Break is controlled by a [PI Controller](https://www.youtube.com/watch?v=VVOi2dbtxC0).
* The model was trained using all tracks of [Udacity self-driving-car-simator](https://github.com/udacity/self-driving-car-sim) _Version 2_ and tested with track 2 of _Version 1_ . The idea was try model with a track that it never saw.
* The model was based to [End-to-End Deep Learning for Self-Driving Cars](https://devblogs.nvidia.com/deep-learning-self-driving-cars/) network arquitecture.
## Test videos
**Test 1**: Test model on track that was used to generate training & test samples.
**Test 2**: Test model over a track that it never saw.
## Requeriments
* [anaconda](https://www.anaconda.com/download/#linux)
* 7z
* A respectable video card (i.e. [GeForce GTX 1060](https://www.nvidia.com/en-us/geforce/products/10series/geforce-gtx-1060/) or higher)
## Setup
**Step 1**: Create project environment.
```bash
$ conda env create --file environment.yml
```
**Step 2**: Activate environment.
```bash
$ conda activate self-driving-car-model
```
**Note**: This step is required before run `train_mode.py` or `drive.py`.
## Train model
You can train and adjust model from [
Self driving car model analysis
](https://github.com/adrianmarino/self-driving-car-model/blob/master/model-analysis.ipynb) notebook or use
`train_model.py` script.
First of all you need a dataset, but already exist a dataset that was created to train the model, so you can download this. To train model follow next steps:
**Step 1**: Activate environment.
**Step 2**: Download dataset from [here](https://drive.google.com/file/d/1O84dTrE2j1J9xhPmlJdVwRJ55WcJlMQN/view?usp=sharing) to project path.
**Step 3**: Extract dataset.
```bash
$ 7z x self-driving-car-dataset.7z
```
**Step 4**: Train model (using `train_model.py` script).
```bash
$ python train_model.py
```
**Notes**
* This script load model last weights from /checkpoints path if it exists.
* Can change epochs and learning rate with `--epochs` value `--lr` value.
**Step 5**: Monitor train/validation loss and steering angle RMSE from [Tensorboard](https://www.tensorflow.org/guide/summaries_and_tensorboard). First of all you need run tensor board as follows:
```bash
$ tensorboard --logdir logs
```
**Note**: `train_model.py` script write metrics under `./logs` directory and tensoboard read from this.
**Step 6**: Go to [http://localhost:6006](http://localhost:6006) url.
## Play model
To test model with a track that it never saw use [self-driving-car-sim](https://github.com/udacity/self-driving-car-sim) _Version 1_. You can also test model with any track from _Version 2_, but the model already know this tracks, given these were used to generate the training and validation samples.
### To play model
**Step 1**: Download [self-driving-car-sim](https://github.com/udacity/self-driving-car-sim) _Version 1_.
**Step 2**: Execute simulator, select _Track 2_ and press _Autonomous Mode_.
**Step 3**: Activate environment.
**Step 4**: Execute model client.
```bash
$ python drive.py checkpoints/weights__loss_0.0525__rmse_0.2055.h5
```
