Decision Process
For the purpose of the semester thesis, the creation of a complete driving simulator with advanced VR or Force-Feedback functionality was out of the discussion. For the hardware, we defined early on the G29 driving wheel as our input device of choice and one or multiple screens for displaying the simulation.
For the selection of a suitable software, it is important that the complexity of the tool is not too high, such that we are able to set up our scenario in a manner of weeks.
Nevertheless, its flexibility needs to be high enough to adjust the environment to our liking.
We need to be able to make the driving scenario repeatable, possibly with programmable interacting with the user. Furthermore, we need access to the user input and states of the vehicle in order to apply the algorithm and lastly, it would be beneficial, if the model of the car (dynamic & kinematic model) is available in a mathematical notation and readily adjustable.
From this high-level requirements, we derived the following five key criteria with subpoints mentioned in the order of importance.
Model:
- Is a description of the model available or do we need a system identification?
- Is it possible to implement our own kinematic & dynamic model in terms of
difference
equations?
Interface:
- Is the data (input and car states) easily accessible and exportable?
- To what amount are we able to adjust the source code? How much effort does it take to do so?
- Is there native support for the G29 joystick? If not, how much overhead does it take?
Scenarios:
- Can I create my own scenarios?
- Does the creation of scenarios scale easily with the complexity of its complexity?
- Is it possible to integrate a programmable car in the scenario?
- Are there additional features such as pedestrian, city environment, weather conditions?
General:
- Which Technology is used and what operating systems are supported?
- Is it open-source? How much would a student license cost?
- Is there an active community or a help centre for support?
Appearance:
- How appealing is the graphical representation?
- How realistic is the driving feeling?
These key criteria were applied to a selection of six software solutions. In the following, we present a short summary for each of the candidate and a visualization by means of a decision matrix.
Torcs:
- opensource, widely used in research (bad in scenario generation)
CarMaker
- commercial (bad free version, with some rather harsh condition on its usage), simulation interface
Vires vtd
- possibly most suitable solution, used in similar scenarios, rather complex framework
A student VTD-license for 6 month costs 500,00 €uro and contains the following modules:
Conclusion
As can be seen from the decision matrix, Torcs and Vires VTD and IPG are the most fitting solutions. But Vires VTD has the big drawback of being a commercial software. As a result, we decided to use the Torcs simulation framework as our first choice. The only real drawback of this solution is the limited scenario generation. That is why we keep the IPG Carmaker as a backup solution (since it's performance is close to Vires VTD but with a free student version) for a possible extension of the scenarios and started communication with ETH lab.