D-robotics Robotic Manipulation Platform

🔥 Latest Work: VO-DP

VO-DP-1080_30.mp4

🌟 Update Log

• 2025-11: The robotwin2.0 simulator is available now! Check out the simulation/robotwin2.0
• 2025-10: We release the code and dataset for our latest work VO-DP: Semantic-Geometric Adaptive Diffusion Policy for Vision-Only Robotic Manipulation.

⚙️ Installation

Basic Environment Setup

git clone https://github.com/D-Robotics-AI-Lab/DRRM.git
cd DRRM
conda create -n drrm python=3.10
conda activate drrm
pip install -e .

VODP Environment Setup

mkdir -p third_party
cd third_party
git clone https://github.com/facebookresearch/vggt.git
cd vggt
pip install .
cd ../..

Robotwin Environment Setup

See simulation/robotwin/README.md for detailed simulator installation and usage steps.

📊 Dataset Preparation

Create the datasets/ directory and download one of the preprocessed dataset bundles from HuggingFace:

mkdir -p datasets

Preprocessed bundles available (choose one):

• drrm_robotwin1.0_D435_200_rgb — RGB-only version (no point clouds)
• drrm_robotwin1.0_D435_200_pcd — includes point clouds

Visit https://huggingface.co/datasets/D-Robotics/DRRM to download the dataset and place it under datasets/.

📑 Training

1. Modify the acceleration configuration file based on your training environment: configs/accelerate_config.yaml
2. In the training script scripts/train_demo.sh, specify the following parameters:
   • dataset: Path to your training dataset
   • task: Your training task
   • demo: Number of demonstration samples to use (set to null for unlimited)
   • config_dir: Training configuration directory (refer to configs/)

• Training VODP

accelerate launch\
    --config_file configs/accelerate_config.yaml \
    main.py \
    --config-path="configs/vodp_train" \
    --config-name="vodp_23d_1f.yaml" \
    train_dataset.path=datasets/lerobot_D435_200 \
    train_dataset.task=block_hammer_beat \
    train_dataset.demo=100

• Training DP

accelerate launch\
    --config_file configs/accelerate_config.yaml \
    main.py \
    --config-path="configs/dp_train" \
    --config-name="dp.yaml" \
    train_dataset.path=datasets/lerobot_D435_200 \
    train_dataset.task=block_hammer_beat \
    train_dataset.demo=100

• Training DP3

accelerate launch\
    --config_file configs/accelerate_config.yaml \
    main.py \
    --config-path="configs/dp3_train" \
    --config-name="dp3.yaml" \
    train_dataset.path=datasets/lerobot43d_D435_200 \
    train_dataset.task=block_hammer_beat \
    train_dataset.demo=100

🤖 Simulation Evaluation

DRRM is compatible with the Robotwin simulator — refer to the following files:

• scripts/eval/eval_robotwin.sh — convenience shell wrapper used for experiments.
• scripts/eval/robotwin_exp/ — example experiment wrappers used in our paper.
• simulation/robotwin/script/ — simulator-facing Python evaluation scripts (e.g. eval_policy_vodp.py, eval_policy_dp.py, eval_policy_dp3.py).

Supported benchmark tasks:

[
    'block_hammer_beat', 'bottle_adjust', 'container_place',
    'dual_bottles_pick_hard', 'put_apple_cabinet',
    'tool_adjust', 'pick_apple_messy', 'dual_bottles_pick_easy',
    'diverse_bottles_pick', 'empty_cup_place', 'shoe_place',
    'dual_shoes_place', 'blocks_stack_easy', 'block_handover'
]

Basic usage

Replace YOUR/CHECKPOINT/DIR and YOUR/SAVE/DIR with the paths to your checkpoint directory and the directory where you want to store evaluation results. The --num-process flag controls parallel simulator workers.

# Evaluate VODP
python simulation/robotwin/script/eval_policy_vodp.py \
    --checkpoint-dir YOUR/CHECKPOINT/DIR \
    --save-dir YOUR/SAVE/DIR \
    --task-name TASK_NAME \
    --num-process 8 \
    --seed 0

# Evaluate DP
python simulation/robotwin/script/eval_policy_dp.py \
    --checkpoint-dir YOUR/CHECKPOINT/DIR \
    --save-dir YOUR/SAVE/DIR \
    --task-name TASK_NAME \
    --num-process 8 \
    --seed 0

# Evaluate DP3
python simulation/robotwin/script/eval_policy_dp3.py \
    --checkpoint-dir YOUR/CHECKPOINT/DIR \
    --save-dir YOUR/SAVE/DIR \
    --task-name TASK_NAME \
    --num-process 8 \
    --seed 0

👏 Citation

@article{ni2025vodp,
  title={VO-DP: Semantic-Geometric Adaptive Diffusion Policy for Vision-Only Robotic Manipulation},
  author={Zehao Ni and Yonghao He and Lingfeng Qian and Jilei Mao and Fa Fu and Wei Sui and Hu Su and Junran Peng and Zhipeng Wang and Bin He},
  journal={arXiv preprint arXiv:2510.15530},
  year={2025}
}