SE(3)-Equivariant Diffusion Policy in Spherical Fourier Space
Authors: Xupeng Zhu, Fan Wang, Robin Walters, Jane Shi
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We conduct simulation experiments using the Mimic Gen (Mandlekar et al., 2023) environment, built on the Mujoco simulator (Todorov et al., 2012)... We further evaluate the performance of SDP across 5 physical tasks...Table 4 presents six ablations... |
| Researcher Affiliation | Collaboration | Xupeng Zhu 1 2 Fan Wang 3 Robin Walters 2 Jane Shi 3 1Work was done as an intern at Amazon Robotics 2Khoury College of Computer Science, Boston, Massachusetts, USA 3Amazon Robotics, Boston, Massachusetts, USA. Correspondence to: Xupeng Zhu <EMAIL>. |
| Pseudocode | No | The paper describes the methodology in text and figures (Figure 2, Figure 3) but does not provide a formal pseudocode or algorithm block. |
| Open Source Code | Yes | Code is available at https: //github.com/amazon-science/ Spherical_Diffusion_Policy. |
| Open Datasets | Yes | We conduct simulation experiments using the Mimic Gen (Mandlekar et al., 2023) environment... We use Gello (Wu et al., 2023) to collect demonstrations... |
| Dataset Splits | No | For the single-arm tasks, we collect 30 successful human demonstrations. Additionally, we record an extra 10% demos as the recovery demos at specific poses. Similarly, for the bi-manual Grasp Box task, we collect 33 demos at random SE(3) poses. For more challenging bi-manual tasks like Flip Book and Pack Package, we collect 66 demos. Each baseline is evaluated on 20 rollouts per physical task, with each rollout initialized using object poses in novel SE(3) poses unseen in the training set. |
| Hardware Specification | No | The paper mentions the robotic workstation components (UR5e manipulators, Real Sense D415 cameras) but does not specify the computational hardware (e.g., exact GPU/CPU models) used for training or inference, beyond a 'commercial GPU with 24GB RAM'. |
| Software Dependencies | No | The paper mentions several software tools and frameworks (e.g., Mujoco simulator, DDPM, DDIM, Gello) but does not provide specific version numbers for any of them or other ancillary software components. |
| Experiment Setup | Yes | Hyperparameters for diffusion-based baseline methods are listed in Table A2. SDP generally adopts Diffusion Policy 's hyperparameters, except for batch size, because SDP is heavier than other baselines. Table A2 includes: Batch Size 32, Prediction Horizon 16, Action Horizon 8, Learning Rate 1e-4, Epochs 500, Learning Rate Scheduler cosine, Noise Scheduler DDPM, Diffusion Train/Test Step 100, Encoded Scene Dimension 128 for SDP. |