For the first year, we are competing in the 2018 edition of the Audi Autonomous Driving Cup (AADC).
The contest consists in developing fully automatic driving functions for a 1:8 model vehicle provided by Audi.
For the mandatory program, each team is supposed to complete a list of manouvres on a simple route specified in openDrive format.
This year's novelty is a merging scenario into the regular lane, as described in the simulation below.
In openDrive format, the route is described by a list of sectors, which can be normal roads or junctions. Normally, the trajectory inside roads refers to the center of the road, while in junctions it refers to the center of the connecting lane the car is following (see the bottom right frame).
The trembling aspect of images is due to a large amount of noise added in the simulation.
Cars are equipped with a linux operating system and the Audi's ADTF development environment, providing access to sensor technology and vehicle actuators.
Among other things, a suitable ADTF module supplies the vehicle position on the route, exploiting both the vehicle odometry (speed signal, 6D acceleration sensor), and synchronisation with a few traffic signs positioned on the track.
The position computed by the ADTF module is not very accurate, and a crucial task is to improve it via the signals recevied from video cameras.
A simple but fast and effective way to do it is to exploit hough-transformations for discovering lines, as in the video below (we used library functions of openCV). In this way, it is possible to compute with high accuracy the distance of the cars from borders and crossings.
Several thesis available on this topic! (simulation, artificial intelligence, computer vision, learning, ...)
We shall eventually organize a new team for next year.
Contact me if interested.