Research map / Vision-language and multimodal / Robot action and navigation
LaMP: Learning Vision-Language-Action Policies with 3D Scene Flow as Latent Motion Prior
Xinkai Wang, Chenyi Wang, Yifu Xu, Mingzhe Ye, Fu-Cheng Zhang, Jialin Tian, Xinyu Zhan, Lifeng Zhu, Cewu Lu, Lixin Yang
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 |
|---|---|---|---|
| add Motion Expert (3D scene flow prior) (ablation) | LaMP · LIBERO | avg success: 92.4 -> 98.3 | Improved |
| add Motion Expert (ablation) | LaMP · LIBERO-Long | success: 78.2 -> 96.7 | Improved |
| add Motion Expert (ablation) | LaMP · SimplerEnv-WidowX | avg success: 56.3 -> 79.2 | Improved |
| add Motion Expert (ablation) | LaMP · LIBERO-Plus OOD | avg success: 71.6 -> 79.3 | Improved |
| use 3D scene flow instead of 2D flow (ablation) | LaMP · SimplerEnv Stack Block | success: 58.3 -> 75.0 | Improved |
| use 3D scene flow instead of 2D flow (ablation) | LaMP · SimplerEnv Put Spoon | success: 70.8 -> 79.1 | Improved |
| use 3D scene flow instead of 2D flow (ablation) | LaMP · SimplerEnv Put Eggplant | success: 91.7 -> 95.8 | No clear effect |
| gated cross-attention instead of additive fusion (ablation) | LaMP · SimplerEnv Stack Block | success: 33.3 -> 75.0 | Improved |
| gated cross-attention instead of Concat+MLP (ablation) | LaMP · SimplerEnv Stack Block | success: 45.8 -> 75.0 | Improved |
| gated cross-attention vs Add and Concat+MLP (ablation) | LaMP · SimplerEnv Put Eggplant | success: 91.7/83.3 -> 95.8 | Improved |
| LaMP vs pi0 and 3D FDP on real robot | LaMP · Real-world in-domain and OOD | task progress: OOD 52.5 -> 62.5 | Improved |
| use one-step partial denoising of motion expert (ablation) | LaMP · Latency | latency/memory: 130.8 -> 177.1 ms | Mixed |
Closest papers by meaning
- PriorVLA: Prior-Preserving Adaptation for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- Beyond World-Frame Action Heads: Motion-Centric Action Frames for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- RotVLA: Rotational Latent Action for Vision-Language-Action Model (NeurIPS 2026 · similarity 0.88)
- ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models (NeurIPS 2026 · similarity 0.88)
- BridgeVLA: Input-Output Alignment for Efficient 3D Manipulation Learning with Vision-Language Models (NeurIPS 2025 · similarity 0.88)
- Beyond Data Scaling: Representation-Centric Pre-training for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.88)
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.
