Trained
DP + Environment States (DPES) checkpoints for the
KinDER
physical-reasoning benchmark (RSS 2026).
Each checkpoint is an imitation learning policy trained from
~100 human
demonstrations per environment.
The training code lives in
kinder-diffusion-policy
(a fork of
diffusion_policy).
Demonstrations are available at
kinder-bench/kinder-datasets.
The environment and robot state vectors are each passed through a small MLP
encoder before concatenation with the visual features, giving the policy direct
access to precise geometric information that may be hard to extract from pixels
alone.
Action chunk predicted by iterative DDPM denoising, identical to DP.
1# Clone and set up kinder-diffusion-policy
2git clone git@github.com:Princeton-Robot-Planning-and-Learning/kinder-diffusion-policy.git
3cd kinder-diffusion-policy
4# Follow the environment setup instructions in the repo README
5mamba activate robodiff
1# Install the kinder-imitation-learning inference utilities
2cd kinder-baselines/kinder-imitation-learning
3uv pip install -r prpl_requirements.txt
4uv pip install -e ".[develop]"
1cd ~/kinder-diffusion-policy
2mamba activate robodiff
3python policy_server.py --ckpt-path /path/to/sweep3d/epoch=0300-train_loss=0.001.ckpt
1cd kinder-baselines/kinder-models/scripts
2python inference.py \
3 --env-name kinder/SweepIntoDrawer3D-o5-v0 \
4 --save-videos \
5 --num-seeds 1 \
6 --num-episodes 5 \
7 --max-steps 200
1# Convert raw teleoperation recordings to HDF5
2cd kinder-baselines/kinder-models/scripts
3python demos_to_hdf5.py \
4 --teleop_data_dir $YOUR_DATA_DIR \
5 --output_path $OUTPUT_HDF5_PATH \
6 --render_images
7
8# Train with the DPES config (includes state inputs)
9cd ~/kinder-diffusion-policy
10mamba activate robodiff
11python train.py --config-name=train_sweep3d_image_state
1@inproceedings{huang2026kinder,
2 title = {KinDER: A Physical Reasoning Benchmark for Robot Learning and Planning},
3 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},
4 booktitle = {Robotics: Science and Systems (RSS)},
5 year = {2026}
6}
