Content:
The goal of this project is to develop a self-driving car in Grand Theft Auto V like in this example:
- Tuto - Python Plays: Grand Theft Auto V
- sentdex Github
- Youtube channel - Python Plays: Grand Theft Auto V
- Self-driving motorbike video
The first challenge is to record the display to replay it for further analysis (like a simple car detection here):
We have 2 options:
The folder ps4_macro contains the controller record and python functions PS4_control.py provides 2 functions to:
- read XML controller file as DataFrame,
- save it as Excel file
PyWin32 to record keyboard
/!\ Map A to Q on French keyboard!!!
1 epoch only with training0.py
Tensorflow on i7-6700HQ CPU @ 2.60GHz / 8Go RAM
21 steps == 6m 27s
I stopped it to train on my GPU instead
Tensorflow on NVIDIA GeFOrce GTX 960M
21 steps == 3m 40s
92 steps == 17m 6s
- Building a Self-Driving Vehicle in GTA-V Using Deep Learning and Convolutional Neural Network
- https://github.com/Alzaib/Autonomous-Self-Driving-Car-GTA-5
- https://medium.com/@alexlau27/playing-ps4-with-tensorflow-2-1-and-alexnet-2aeec0e2a5d8
- https://github.com/komefai/PS4Macro
- https://github.com/starshinata/PS4-Keyboard-and-Mouse-Adapter
- https://bitbucket.org/visinf/projects-2016-playing-for-data/src/master/
- https://arxiv.org/abs/1608.02192
- http://download.visinf.tu-darmstadt.de/data/from_games/data/eccv-2016-richter-playing_for_data.pdf
| GTA driving data | semantic sampling |
|---|---|
Precise Synthetic Image and LiDAR (PreSIL) Dataset from Waterloo Intelligent Systems Engineering Lab
Linearized Depth Map (Grey and Colour)
Object Instance Segmentation
