TORCS autonomous driving

This repository contains source code to train and drive a car in TORCS by itself.

Dependencies

TORCS-ROS, TORCS, ROS Kinetic
python2: nengo, keras, tensorflow, numpy, opencv
python3: keras, tensorflow, numpy, opencv

How to run

1. start torcs: torcs
2. configure the desired track, choose scr_server as driver
3. run: roslaunch torcs_ros_bringup torcs_ros.launch rviz:=false driver:=false
4. go into the nengo_controller folder and run: python2 controller.py

Structure

The nengo_controller folder contains the code needed to drive the car based on all the given sensor values. The folder src/collect_img_sensor_data contains a ROS node to collect training data for the DNN. The folder src/train-deep-neural-network contians code to train a deep neural network to infer angle and car displacement from a driver's view input image.

├── final-presentation-complete
│   └── Bilder
├── nengo_controller
│   ├── data
│   │   └── processed_data
│   └── nengo_ros
├── report
│   ├── attachments
│   └── paper
└── src
    ├── collect_img_sensor_data
    │   ├── data-aalborg-2laps-640x480
    │   ├── data-alpine_1-2laps-640x480
    │   ├── data-alpine_2-2laps-640x480
    │   ├── data-brondehach-2laps-640x480
    │   ├── data-cg_speedway_1-2laps-640x480
    │   ├── data-cg_track_2-2laps-640x480
    │   ├── data-cg_track_3-2laps-640x480
    │   ├── data-cg_track_3-2laps-640x480-1sthood
    │   ├── data-cg_track_3-2laps-640x480-3rdclose
    │   ├── data-cg_track_3-2laps-640x480-3rdfar
    │   ├── data-corkscrew-2laps-640x480
    │   ├── data-e_road-2laps-640x480
    │   ├── data-etrack_1-2laps-640x480
    │   ├── data-etrack_2-2laps-640x480
    │   ├── data-etrack_3-2laps-640x480
    │   ├── data-etrack_4-2laps-640x480
    │   ├── data-etrack_6-2laps-640x480
    │   ├── data-forza-2laps-640x480
    │   ├── data-olethros_road_1-2laps-640x480
    │   ├── data-ruudskogen-2laps-640x480
    │   ├── data-street_1-2laps-640x480
    │   ├── data-wheel_1-2laps-640x480
    │   ├── data-wheel_2-2laps-640x480
    │   ├── launch
    │   └── src
    └── train-deep-neural-network