Research map / Image generation / Autoregressive and latent generation
EDGI: Equivariant Diffusion for Planning with Embodied Agents
Johann Brehmer, Joey Bose, Pim De Haan, Taco Cohen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use SE(3)xZxS_n-equivariant architecture instead of Diffuser | Diffuser · Kuka unconditional stacking | reward: 59.7 -> 68.7 | Improved |
| use equivariant architecture instead of Diffuser | Diffuser · Kuka conditional stacking | reward: 46.0 -> 52.0 | Improved |
| use equivariant architecture instead of Diffuser | Diffuser · Kuka rearrangement | reward: 49.2 -> 47.2 | No clear effect |
| use equivariant architecture instead of Diffuser | Diffuser · Navigation (standard) | reward: 94.9 -> 95.1 | No clear effect |
| use equivariant architecture on restricted-orientation training data | Diffuser · Navigation SO(3) generalization | reward: 5.6 -> 83.3 | Improved |
| use equivariant architecture on restricted-orientation training data | Diffuser · Kuka unconditional SO(3) generalization | reward: 38.7 -> 69.0 | Improved |
| use equivariant architecture on restricted-orientation training data | Diffuser · Kuka conditional SO(3) generalization | reward: 16.7 -> 35.9 | Improved |
| use equivariant architecture on restricted-orientation training data | Diffuser · Kuka rearrangement SO(3) generalization | reward: 17.8 -> 45.0 | Improved |
| diffusion-based planners vs offline RL baseline CQL | Diffuser · Kuka average | reward: 8.1 -> 51.6 | Improved |
| train on 10% of data (Kuka) or 0.01% (navigation) (ablation) | EDGI · sample efficiency | reward | No clear effect |
| add SO(3) data augmentation to Diffuser training (ablation) | Diffuser · sample efficiency | reward | Improved |
| equivariant to only SE(3) or only S_n instead of full group (ablation) | EDGI · sample efficiency | reward | Mixed |
Closest papers by meaning
- ET-SEED: EFFICIENT TRAJECTORY-LEVEL SE(3) EQUIVARIANT DIFFUSION POLICY (ICLR 2025 · similarity 0.88)
- A Practical Guide for Incorporating Symmetry in Diffusion Policy (NeurIPS 2025 · similarity 0.87)
- 3D Equivariant Visuomotor Policy Learning via Spherical Projection (NeurIPS 2025 · similarity 0.86)
- Simple Hierarchical Planning with Diffusion (ICLR 2024 · similarity 0.85)
- Boosting Sample Efficiency and Generalization in Multi-agent Reinforcement Learning via Equivariance (NeurIPS 2024 · similarity 0.85)
- Prior-Guided Diffusion Planning for Offline Reinforcement Learning (NeurIPS 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
