kinder-bench/kinder-datasets

KinDER Demonstration Datasets

Human-collected robot demonstration datasets for the KinDER physical-reasoning benchmark (RSS 2026).

These datasets are used to train the imitation learning baselines (kinder-imitation-learning) and the model-based RL baseline (kinder-mbrl) shipped with KinDER.

---

Dataset summary

File	KinDER environment	Episodes	Total steps	Mean ep. length
motion2d_p0.hdf5	Motion2D-p0	111	4 634	41.7
stickbutton2d_b1.hdf5	StickButton2D-b1	114	8 141	71.4
dynobstruction2d_o1.hdf5	DynObstruction2D-o1	101	8 901	88.1
dynpushpullhook2d_o5.hdf5	DynPushPullHook2D-o5	103	22 435	217.8
BaseMotion3D_110.hdf5	BaseMotion3D	110	2 785	25.3
shelf3d_106_new.hdf5	Shelf3D	106	46 598	439.6
sweep3d_100.hdf5	SweepIntoDrawer3D	100	51 741	517.4
transport3D_o2.hdf5	Transport3D-o2	117	126 860	1 084.3
Total		862	272 095

---

Data collection

2D environments (motion2d, stickbutton2d, dynobstruction2d, dynpushpullhook2d): Demonstrations were collected using a PS5 controller.

3D environments (BaseMotion3D, shelf3d, sweep3d, transport3D): Demonstrations were collected using an iPhone web app (XR Browser) or a VR headset, streaming 6-DoF pose commands to the TidyBot robot in simulation.

---

File format

Every .hdf5 file follows the same schema:

data/
  demo_0/
    actions       (T, action_dim)   — control commands applied at each step
    obs/
      robot_state (T, robot_dim)    — proprioceptive robot state
      env_state   (T, env_dim)      — environment / object state
      image       (T, 224, 224, 3)  — top-down / scene RGB image  [2D envs]
      base_image  (T, 224, 224, 3)  — robot base camera           [3D envs]
      wrist_image (T, 224, 224, 3)  — robot wrist camera          [3D envs]
      overview_image (T, 224, 224, 3) — static overview camera    [3D envs]
  demo_1/
    ...

All images are uint8 RGB arrays of shape (T, 224, 224, 3).
State and action arrays are float32.

Per-dataset dimensions

2D environments

File	action_dim	robot_dim	env_dim	Cameras
motion2d_p0.hdf5	5	9	10	image
stickbutton2d_b1.hdf5	5	9	19	image
dynobstruction2d_o1.hdf5	5	24	44	image
dynpushpullhook2d_o5.hdf5	5	24	106	image

3D environments (TidyBot)

File	action_dim	robot_dim	env_dim	Cameras
BaseMotion3D_110.hdf5	11	19	3	base, wrist, overview
shelf3d_106_new.hdf5	11	22	23	base, wrist, overview
sweep3d_100.hdf5	11	22	131	base, wrist, overview
transport3D_o2.hdf5	11	19	48	base, wrist, overview

3D action space (11-dim): [base_x, base_y, base_yaw, joint_1, …, joint_7, gripper] — delta joint-position commands.

3D robot state (19–22-dim): base pose + joint angles + joint velocities + gripper state.

---

Loading the data

With h5py (any baseline)

import h5py
import numpy as np

with h5py.File("sweep3d_100.hdf5", "r") as f:
    episodes = list(f["data"].keys())           # ["demo_0", "demo_1", ...]
    ep = f["data"][episodes[0]]
    actions     = ep["actions"][:]              # (T, 11)
    robot_state = ep["obs"]["robot_state"][:]   # (T, 22)
    env_state   = ep["obs"]["env_state"][:]     # (T, 131)
    base_image  = ep["obs"]["base_image"][:]    # (T, 224, 224, 3)

With kinder-mbrl (world model training)

python experiments/train_world_model.py \
    --mode train \
    --hdf5_path sweep3d_100.hdf5 \
    --output_dir output \
    --epochs 1000

The training script reads robot_state and env_state from obs/ and computes per-step (state, action, delta) transitions automatically.

With kinder-imitation-learning (Diffusion Policy)

Convert raw demos to the diffusion policy format:

cd kinder-baselines/kinder-models/scripts
python demos_to_hdf5.py \
    --teleop_data_dir $YOUR_DATA_DIR \
    --output_path $OUTPUT_HDF5_PATH \
    --render_images

Then train:

cd ~/kinder-diffusion-policy
mamba activate robodiff
python train.py --config-name=train_sweep3d_image

---

Baselines trained on this data

Baseline	Reference
Diffusion Policy (DP)	kinder-imitation-learning
DP + Environment States (DPES)	kinder-imitation-learning
Finetuned π0.5 VLA	kinder-openpi
MLP World Model + Random-shooting MPC	kinder-mbrl

---

Citation

If you use these datasets, please cite the KinDER benchmark:

@inproceedings{huang2026kinder,
  title     = {KinDER: A Physical Reasoning Benchmark for Robot Learning and Planning},
  author    = {Huang, Yixuan and Li, Bowen and Saxena, Vaibhav and Liang, Yichao and Mishra, Utkarsh and Ji, Liang and Zha, Lihan and Wu, Jimmy and Kumar, Nishanth and Scherer, Sebastian and Xu, Danfei and Silver, Tom},
  booktitle = {Robotics: Science and Systems (RSS)},
  year      = {2026}
}