Visualization

DI-drive builds an online visualizer for the user to check the situation of a running simulator and save gif/video files. The visualizer is added in the environment’s render. When calling this method, if a visualizer is created by config settings, it will take a sensor data as canvas and put some info text onto it, including actions, road options, agent state, reward, traffic light records and so on. Then the canvas can be showed in the screen or saved as a gif or video file to watch. A visualizer can be customized by config.

visualizer = dict(
    type=None,
    show_text=True,
    outputs=list(),
    save_dir='',
    frame_skip=0,
)

• type

  The observation tag in simulator config’s obs key. Visualizer use it as a canvas image.

• show_text

  Whether to add text above the image. Text information is found in observation or action dict.

• outputs

  Output types of visualizer, can be one or more items of [‘show’, ‘gif’, ‘video’]

• save_dir

  Folder path to save visualized files.

• frame_skip

  The number of frames skipped between two steps when visualizing. For example, if set to 2, the painting and showing will be run every 3 frames.

Here we show some visualization results. You can check how to get these visualizations in auto run

Visualization of RGB camera

Visualization of Bird-eye View image

There is also another visualization method using gym.wrappers.Monitor to record the visualized canvas returned by render method of an environment, and make video using ffmpeg lib. This can be switched on by calling enable_save_replay of an env manager. The visualization video and logs will be recorded in the provided folder path.

my_env = BaseEnvManager(env_fn, cfg.enb.manager)
my_env.enable_save_replay(cfg.env.replay_path)

Note

If you are using your own env class, be careful dealing with env instance when using visualization with ‘enable_save_replay’.