Section 1 Lap Time Challenge - ika-rwth-aachen/acdc GitHub Wiki

The main practical task of this challenge is to implement your own vehicle controller as described here. We are making a challenge out of it where you can participate.

In particular:

• Implement an improved longitudinal control rule in computeTargetVelocity()
• Implement an improved lateral control rule in computeTargetSteeringAngle()

Try to achieve at least one lap without major wall crashes. Be creative and use any sort of scaling factors or mathematical functions that help you implement a fast and smooth vehicle behavior. You can also have a look at the background information.

⚠️ Do not forget to save your source files and build your source code in between making changes and trying them out:

catkin build racing --no-deps  # only build racing package (faster)

After that you can start again the simulation environment as well as the vehicle controller with the already known commands:

roslaunch racing flatland_simulation.launch
roslaunch racing racing_controller.launch

• The lidar sensor provides five data points where our current implementation only uses three of them. Consider also using more sensor data which could naturally improve the resulting driving behavior.

• Base the longitudinal control on the front measurement, and the lateral control on the left and right measurements.

To validate your submission, it should consist of:

• must: An .mp4 or .gif animation as "proof" (see header gif above)
  • must cover exactly one lap (+ roughly 3s before and after)
  • must show RViz with the cart driving through the track
  • must also show the terminal output of the lap time measurement
    • The fastest out of the latest 5 shown lap times is taken.
  • can be recorded using e.g. SimpleScreenRecorder
    • sudo apt install simplescreenrecorder to install it
    • simplescreenrecorder to start it
    • please keep the file size reasonable
• We recommend to include your edX username so that we can honor you on the leaderboard.
• It would be really interesting to share also your source code and maybe contribute to a combined solution at the end of this course.
• Optional: To better compare submissions across the leaderboard make sure to include information on the number of lidar points used in your submission. For a more advanced challenge, try to change the number of reflection points and try to beat the fastest lap with fewer or more points.

⚠️ Compress all files to one single ZIP file and upload it to the Sciebo project box here. We highly recommend that your file name also contains your edx username.

Note: fastest students from last years: 9.37 seconds 🚗 🚗

• We are aware of a number of possibilities to cheat in this submission.
• It would be too much effort to build a cheat-proof submission system as of yet.
• We will exclude any obvious cheats from the leaderboard.
• If you take someone else's code and improve it only a little bit, then please be fair and acknowledge the original developer in your submission.
• Cheaters lose on the moral leaderboard.

If you are interested in more serious challenges: As of 2020, the CARLA Leaderboard ranks different algorithms for automated vehicle guidance. Besides this, there are numerous ongoing challenges for environment perception tasks, e.g. KITTI, Cityscapes, or nuScenes.