Specific data information

[rohjunha]

Specific data information

[rohjunha]

Hi @kaiiwoo, I think we have discussed this in the email thread. You can check how the data is parsed from https://github.com/rohjunha/language-grounded-driving/blob/master/data/types.py#L150. This is a concatenation of the state of the car, a control of the car, a waypoint id, and road options that are related to the labels of the waypoint.
Yes, OfflineGeneratorEnvironment should generate data.

I believe the redundant expression indicates cases where their actions are repeated twice. The ordinal expression means expressions that contain numbers such as first turn or second turn.

data.mdb contains raw driving data. segment.json has information to transform raw data into low-level and high-level segments. Low-level segments contain unit actions and are defined by multiple DriveData. High-level segments define sentence-level behavior including multiple low-level segments. For instance, high-level turning behavior includes going straight to the intersection, turning at the intersection, and (optionally) stopping after the turn. In this example, a single high-level segment 'left' will have three low-level segments with the labels of 'lanefollow', 'left', and 'stop'. second_turn.json contains indices of high-level segments to properly assign ordinal sentences when they are used in the evaluation.