chess_bench / README.md
lukasskellijs's picture
upload README.md
7875147 verified
|
Raw
History Blame Contribute Delete
2.35 kB
metadata
license: apache-2.0
task_categories:
  - robotics
tags:
  - LeRobot
  - robotics
  - isaac-sim
  - chess
  - manipulation
configs:
  - config_name: default
    data_files: data/*/*.parquet

This dataset was created using LeRobot.

chess_bench

General chess manipulation demonstrations: a privileged scripted IK expert in NVIDIA Isaac Lab (Arena) picks up any piece from the standard opening and carries it to a random central square. Actions are the native DROID joint-position command (7 absolute arm joint targets + binary gripper), so pi0.5/openpi/GR00T consume them directly. Each episode is labelled with its own instruction (move the {color} {type} from {src} to {dst}); rollouts are kept whether or not they succeed (per-episode success flag preserved in the source HDF5).

  • Robot: Franka Panda + Robotiq 2F-85 (DROID platform); expert drives a DifferentialIKController (dls) into stiff PD joint-position control.
  • Control / record rate: 15 Hz.
  • Episodes: 269 (269 scored success by the privileged check).

Demonstrations per piece

piece episodes
black bishop 17
black king 16
black knight 17
black pawn 43
black queen 12
black rook 12
white bishop 27
white king 4
white knight 28
white pawn 51
white queen 12
white rook 30

Observation / action / state layout

key dtype shape meaning / units
observation.images.front video (H.264, CRF 15) (224,224,3) external RGB, uint8
observation.images.wrist video (H.264, CRF 15) (224,224,3) eye-in-hand RGB, uint8
observation.state float32 (8,) joint_position (7, rad) + gripper_position (1, 0=open 1=closed)
action float32 (8,) 7 absolute arm joint-position targets (rad) + gripper (1: 1=close, 0=open)