English | 中文
Step 1: Collect data from the Simulator
Step 2: Design neural network class (torch.nn.Module)
Write data processing class (torch.utils.data.Dataset)
Step 3: Train the model and save weights (torch.utils.data.DataLoader, optimizer, loss function, …)
Step 4: Drive a car in the Simulator using the trained model
Please open the "Simulator_train" application on your operating system.
Once opened, the Scene will initiate the recording of training data, which will be saved in the "Data/DatasetX/IMG" path and the "Data/DatasetX/VehicleData.txt" file.
In this VehicleData.txt
the data is: the path of IMG, throttle(range: 0 to 1), brakes(range: 0 to 1), sterring(range: -1 to 1).
Please note that you should not remove the "Index.txt" file from the "Data/" directory.
Please open the "Simulator_test" application on your operating system.
and then run your:
python drive.py
In this scene, 3 images are output to your net, and 3 elements of control are acepted.(throttle, brakes and steering)
Function | Keyboard Defaults | Gamepad Defaults |
---|---|---|
Sterring | A/D | Left Stick - Left/Right |
Throttle | W | Left Stick - Up, Right Trigger |
Brakes | S | Left Stick - Down, Left Trigger |
Reset the Position | F1 | - |
Switch mouse control | Mouse right button | - |
Quit | ESC | - |
LRZ Sync+Share https://syncandshare.lrz.de/getlink/fiEQ6uWeJQFmosTX7HsxwG/
Google Drive https://drive.google.com/drive/folders/1YPlSq4V3ArtOUVH11h-doQ2UqCsvQKtQ?usp=sharing
Chair of Robotics, Artificial Intelligence and Real-time Systems · Technical University of Munich
Prof. Dr.-Ing. habil. Alois Christian Knoll
Gao Yichao | Yu Zhou | Michael Sodamin | ||
Song Yinglei | Liu Hao | Liu Hongshen | Liu Lian | Liu Zhuo |
Xie Tongtong | Yang Xiaohan |
周立国 Zhou Liguo