Is it possible to return observations that a well equipped driverless car would receive (e.g. lidar, sonar, multiple cameras, IMU, etc)?
For instance, they could have gone with TORCS:
...or Stunt Rally:
http://stuntrally.tuxfamily.org/
...or Trigger Rally:
http://trigger-rally.sourceforge.net/
(i'm not sure - but Stunt Rally and Trigger Rally may be from the same Trigger codebase)
Trigger Rally also has a webgl version:
https://github.com/CodeArtemis/TriggerRally
All of this is just to say that if it were wanted, a completely open-source driving AI simulator could be easily made.