Research map / Vision-language and multimodal / Robot action and navigation
The Lie We Tell: Correcting the Euclidean Fallacy in Vision Language Action Policies via Score Matching on Tangent Space
Bing-Cheng Chuang, I-Hsuan Chu, Bor-Jiun Lin, YuanFu Yang, Min Sun, Chun-Yi Lee
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 |
|---|---|---|---|
| replace Euclidean diffusion with Lie group diffusion on SE(3) plus GAT encoder | 3D Diffuser Actor · CALVIN ABC->D | average task length: 3.27 -> 3.512 | Improved |
| add GAT encoder (keeping Euclidean diffusion) (ablation) | 3D Diffuser Actor · CALVIN ABC->D | average task length: 3.27 -> 3.474 | Improved |
| add Lie space diffusion (without GAT encoder) (ablation) | 3D Diffuser Actor · CALVIN ABC->D | average task length: 3.27 -> 3.368 | Improved |
| add Lie space diffusion to the GAT-encoder model (ablation) | Lie Diffuser Actor · CALVIN ABC->D | average task length: 3.474 -> 3.512 | Improved |
| add GAT encoder and Lie diffusion | 3D Diffuser Actor · CALVIN ABCD->D | average task length: 3.288 -> 3.584 | Improved |
| add GAT encoder only (Euclidean diffusion, 400K) (ablation) | 3D Diffuser Actor · CALVIN ABCD->D | average task length: 3.288 -> 3.239 | Hurt |
| add Lie diffusion without GAT encoder (ablation) | 3D Diffuser Actor · CALVIN ABCD->D | average task length: 3.288 -> 3.404 | Improved |
| add SE(3) score matching | OpenVLA-OFT · LIBERO Long | success rate: 92.20 -> 94.13 | Improved |
| use Lie diffusion on real robot | 3D Diffuser Actor · Real robot: Sort Blocks | success rate (%): 55 -> 75 | Improved |
| use Lie diffusion on real robot | 3D Diffuser Actor · Real robot: Put Block in Box | success rate (%): 80 -> 75 | No clear effect |
| predict in so(3) and map via exponential map instead of SVD orthogonalization | 3D Diffuser Actor · Orthogonality violation, median | orthogonality violation: 5.7% reduction at median | Improved |
| diffuse on Lie group instead of Euclidean space | 3D Diffuser Actor · Geodesic jitter during denoising | geodesic jitter: nearly an order of magnitude lower | Improved |
Closest papers by meaning
- E0: Expressive Fine-Grained Discrete Action Prediction for Vision-Language-Action Models via Tweedie Discrete Diffusion (NeurIPS 2026 · similarity 0.89)
- LAST: Bridging Vision-Language and Action Manifolds via Gromov-Wasserstein Alignment (ICML 2026 · similarity 0.89)
- Beyond World-Frame Action Heads: Motion-Centric Action Frames for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.88)
- Implicit Drifting Policy: One-Step Action Generation via Conditional Expert Geometry (NeurIPS 2026 · similarity 0.87)
- AffordDP: Generalizable Diffusion Policy with Transferable Affordance (CVPR 2025 · similarity 0.87)
- PDFactor: Learning Tri-Perspective View Policy Diffusion Field for Multi-Task Robotic Manipulation (CVPR 2025 · similarity 0.87)
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.
