# eval.yaml for VLABench/vlabench_primitive_ft_lerobot_video

name: VLABench Primitive Tasks
description: >
  VLABench primitive-tasks benchmark dataset in LeRobot v3.0 format (video-
  compressed visual observations). 5000 episodes / 575,101 frames on a Panda
  robot at 10 fps, covering 10 primitive manipulation tasks. Models trained on
  this dataset can be further evaluated on the generalization tracks in
  VLABench.
evaluation_framework: none

tasks:
  - id: overall
    config: default
    split: overall

  - id: add_condiment
    config: add_condiment
    split: train
    description: Add-condiment manipulation task.

  - id: insert_flower
    config: insert_flower
    split: train
    description: Insert-flower manipulation task.

  - id: select_book
    config: select_book
    split: train
    description: Select-book manipulation task.

  - id: select_chemistry_tube
    config: select_chemistry_tube
    split: train
    description: Select-chemistry-tube manipulation task.

  - id: select_drink
    config: select_drink
    split: train
    description: Select-drink manipulation task.

  - id: select_fruit
    config: select_fruit
    split: train
    description: Select-fruit manipulation task.

  - id: select_mahjong
    config: select_mahjong
    split: train
    description: Select-mahjong manipulation task.

  - id: select_painting
    config: select_painting
    split: train
    description: Select-painting manipulation task.

  - id: select_poker
    config: select_poker
    split: train
    description: Select-poker manipulation task.

  - id: select_toy
    config: select_toy
    split: train
    description: Select-toy manipulation task.