README.md

metadata

pretty_name: G29-HumanDrive-CARLA-5Runs
task_categories:
  - image-segmentation
  - robotics
  - reinforcement-learning
task_ids:
  - semantic-segmentation
size_categories:
  - 10K<n<100K
license: cc-by-4.0
multilinguality: not-applicable
annotations_creators:
  - machine-generated
source_datasets:
  - original
dataset_info:
  features:
    - name: rgb
      dtype: image
    - name: seg
      dtype: image
    - name: steer
      dtype: float32
    - name: throttle
      dtype: float32
    - name: brake
      dtype: float32
    - name: is_dagger
      dtype: int8
    - name: town
      dtype: string
    - name: run
      dtype: string
    - name: frame_id
      dtype: int32
    - name: timestamp_s
      dtype: float32

G29-HumanDrive-CARLA-5Runs

Dataset Description

This dataset contains 70,308 samples collected in CARLA using a Logitech G29 steering wheel.

It includes one representative run from each town:

• runs_01 → Town01
• runs_02 → Town03
• runs_03 → Town04
• runs_04 → Town05
• runs_05 → Town07

Each sample contains:

• RGB image embedded in JPEG
• semantic segmentation image embedded in PNG
• steering command
• throttle command
• brake command
• DAgger flag
• town label
• run identifier
• frame identifier
• timestamp in seconds

The dataset preserves the original image resolution.