Notes

The initial choice was a "Smart" that was adjusted for simulation purposes and equipped with a computer running a simulator software, a bridge, and a monitor in front of the car. The whole system is located in a lab in Garching Hochbrück. The screen in front of the car shows the simulation once it is running. The bridge, consisting of a box underneath the car and a cable system in the car, provides a link between the car and the computer. It is modified in a way that the CAN-Bus' information is directly fed to a computer. The transmitted stream contains every relevant and occurring information e.g. state space, steering angle, speed, velocity, etc. The computer uses a software called "Virtual Test Driving" (VTD) by "Vires" to communicate with the car and simulate a specific scenario. The given information stream is used to build the simulation and grant a realistic behavior. This is enhanced by a real time reaction to inputs given by a driver. Thus steering the car or accelerating feels as real as possible when sitting in a standing car.\
Although the technology is present we nevertheless had some issues with the system. Starting the simulation resulted to be more complicated than anticipated. While 3 different versions of the VTD, the versions 0.88, 1.1, and 1.4, were installed on the computer, only the VTD version 1.1 was working properly. To start a simulation the following chain of steps had to be done, assuming every system (PC, Car, Bridge) is running correctly:\
\
- Execute the commands written in "boot.txt" located on the desktop\
- Open vtd1.1\
- Select "Files" on the left \
- Select "Scenario" and click once on the default scenario\
- Click the "Init" button in the menu bar atop (left to the "Play" button)\
- Go to "Components" and select "load components"\
- Wait for about 30 seconds, tab back to the GUI and press "Inti" again\
- Tab out again and press "Play"
- The simulation should now work properly

However it is important to note, that the steering wheel is calibrated at the beginning of the initial vtd1.1 start. If you move the wheel while going through these steps, the calibration will be lost and the angles might be off (wheel will also not be centered anymore when just driving straight). \

The main problem however was the extraction of specific data from the program. While we managed to get frequent package updates from the running simulation, it was not possible to single out e.g. only the steering angle. Although the information was present, the package was way too big to figure out which exact entry the searched for data is. Unfortunately the documentation for "VTD" was not helpful at all as the two different documents were for older versions and didn't contain the information needed. \
Another problem was the saving of pictures / video. While you can set this up in the simulators GUI, the file is saved as ".mpg" regardless of the settings prior. We tried to change the format to ".tar" and all other displayed options but the file was later on always saved as ".mpg". Unfortunately this was not compatible with our system so far so that we couldn't use this. \

An additional problem arose with the recording and the behaviour of the simulation. Recording with 30fps resulted in a smooth video and thus lots of images to train our neural network with, but it used up a huge amount of storage to a point where we a few seconds would already be about 1GB. Since a lot of images gained with 30fps would be very similar to others as the rate of change is much slower, the training would also be less efficient. Thus we decided to go down in fps to save space and have more significant data. Unfortunately the change also changed the rate in which the simulation ran resulting in an unpleasent 10-5fps driving experience. As driving closely to reality was almost impossible this way, we decided to change our initial idea. As one of the first mentioned options was TORCS, we went with this program for further testing.\