Accepted to Generative Models for Robot Learning Workshop @ ICLR 2025
PDF | arXiv | ICLR Genbot Homepage
Jiahang Cao, Qiang Zhang, Hanzhong Guo, Jiaxu Wang, Hao Cheng, Renjing Xu.
HKUSTGZ, Beijing Innovation Center of Humanoid Robotics, HKU
We introduce a novel policy composition approach, Modality-Composable Diffusion Policy (MCDP), which composes distributional scores from multiple pre-trained diffusion policies (DPs) based on single visual modalities, enabling significant performance improvement without the need for additional training.
Note: This repository as well as the following guidelines are based on RoboTwin, thanks for the authors' open resoure which greatly contributes to the community.
Please carefully follow the guidelines in RoboTwin for installation and data generation.
Running the following command will first search for a random seed for the target collection quantity (default is 100), and then replay the seed to collect data.
bash run_task.sh ${task_name} ${gpu_id}
The DP code can be found in policy/Diffusion-Policy.
Process Data for DP training:
python script/pkl2zarr_dp.py ${task_name} ${head_camera_type} ${expert_data_num}
# As example: python script/pkl2zarr_dp.py dual_bottles_pick_hard L515 100, which indicates preprocessing of 100 dual_bottles_pick_hard task trajectory data using L515 camera.
Then, move to policy/Diffusion-Policy first, and run the following code to train DP:
bash train.sh ${task_name} ${head_camera_type} ${expert_data_num} ${seed} ${gpu_id}
# As example: bash train.sh dual_bottles_pick_hard L515 100 0 0
The DP3 code can be found in policy/3D-Diffusion-Policy.
Process Data for DP3 training:
python script/pkl2zarr_dp3.py ${task_name} ${head_camera_type} ${expert_data_num}
# As example: python script/pkl2zarr_dp3.py dual_bottles_pick_hard L515 100
Then, move to policy/3D-Diffusion-Policy first, and run the following code to train DP3:
bash train_ddpm.sh ${task_name} ${head_camera_type} ${expert_data_num} ${seed} ${gpu_id}
# As example: bash train_ddpm.sh dual_bottles_pick_hard L515 100 0 0
Note: One key difference between our training settings and RoboTwin's default configuration is that we need to align the action distribution space. This means that the noise scheduler for both DP and DP3 needs to be consistent, including aspects such as prediction type and inference time steps. To achieve this, we align DP3's noise scheduler with DP's default DDPM scheduler (in principle, as long as both schedulers are consistent, this is not limited to DDPM). This ensures that the action spaces of both policies remain aligned during inference sampling.
Move to policy/3D-Diffusion-Policy first, then run the following code to evaluate MCDP for a specific task for 100 times:
bash eval_composed.sh ${task_name} ${head_camera_type} ${expert_data_num} ${checkpoint_num} ${seed} ${gpu_id} ${dp_w} ${dp3_w}
# As example: # bash eval_composed.sh dual_bottles_pick_hard L515 100 3000 0 0 0.3 0.7
For each task, we evaluated the baseline performance under L515 (320×180 resolution, FOV 45°) camera. The DP_img and DP_pcd correspond to the diffusion policy based on RGB images (i.e.,, DP) and point cloud (i.e.,, DP3), respectively. We reproduced the DP and DP3 (without using point cloud color) with random seed 0.
- When both DPs have moderate accuracy (both 30%), MCDP often achieves higher accuracy under appropriate weight configurations compared to unimodal DPs.
- When one DP has significantly lower accuracy, MCDP struggles to surpass the highest accuracy of the better-performing unimodal DP.
- Building on Finding 1, the improvement of MCDP is maximized when the better-performing unimodal DP holds a larger weight in MCDP.
Our code is generally built upon: RoboTwin, 3D Diffusion Poliy, Diffusion Policy. We thank all these authors for their nicely open sourced code and their great contributions to the community.
@article{cao2025MCDP,
title={Modality-Composable Diffusion Policy via Inference-Time Distribution-level Composition},
author={Cao, Jiahang and Zhang, Qiang and Guo, Hanzhong and Wang, Jiaxu and Cheng, Hao and Xu, Renjing},
journal={arXiv preprint arXiv:2503.12466},
year={2025}
}
This repository is released under the MIT license. See LICENSE for additional details.