Berkeley Major Map

Version 1.2

Berkeley ROAR team has created a customized CARLA map to facilitate the research and competition using the ROAR software in a simulated mixed urban/suburban environment.

The Berkeley Major Map was created using CARLA and Unreal Engine, and was based on a 3D LIDAR scan of the Berkeley campus and its surrounding hill roads. The map includes a closed circuit that can be driven from both directions. The 3D model of the map accurately represents the altitude variation of the real streets where the model is based on in the real world.

In addition to the provided CARLA server build that is pre-loaded with the Berkeley Major Map model, an official waypoint list file and an orthographic occupancy map file are also provided for reference. ROAR users may freely modify these reference files for the purpose of improving the performance of the autonomous driving agent.

ROAR Server that runs Berkeley Major Map:

Official Waypoints file:

The waypoint file data were recorded from a human driver manually driving a virtual car in the CARLA environment. The human driver ensured no collision of the waypoints with the track borders. However, users of this waypoint file should not expect the resulting performance following this waypoint list is optimal. The users are expected to generate their improved waypoints based on this baseline version to have an edge in ROAR competition.

Occupancy grid map file:

Authors: Jazzy Rao, Xuan Liu, Michael Wu, Tianlun Zhang, Jeffery Shen, and Allen Yang.