Let's build up a list of interesting scenarios to push the driving capabilities of a Torcs Robot.
All of this can be done independently of the vision side of things.
Some inspiration: http://www.darpa.mil/grandchallenge/rules.asp http://www.darpa.mil/grandchallenge/docs/Urban_Challenge_Event_Guidelines_101007.pdf http://www.darpa.mil/grandchallenge/docs/RNDF_MDF_Formats_031407.pdf http://www.darpa.mil/grandchallenge/docs/Technical_Evaluation_Criteria_031607.pdf
CA Driving handbook: http://www.dmv.ca.gov/pubs/dl600.pdf
In general, everything we do will be a race! Even when we obey the speed limit, we'd like the car to pass slower cars, take the shortest route, etc.
General & Racing
The Torcs robot should understand the concept of segments.
- You can go faster around curves with an incline.
- Deal with segments having different friction coefficients.
- Skid around a corner to be faster. (Can AWD cars do that?)
- Deal with not having a pre-calced line. (The track details are not known in advance)
- Make sure you don't jump off a cliff (What is on the other side of that hill?)
- Dealing best with unknown obstacles that just appear. (pothole, deer)
Segments should have properties such as one-way. Segments can have positioned annotations such as stop sign.
Have the car go through a list of waypoints. Have a car go to a list of places and do it with the least amount of driving.
Have a car enter a segment free zone, and find a parking spot.
Once the major rules are in place, then we create scenarios of other cars not following the rules: Create cars that blow through Stop signs. Learning to pass slow cars depending Handling the case where there is an accident and the proper routes are blocked.