Research map / Image generation / Fast diffusion sampling
VITA: Vision-to-Action Flow Matching Policy
Dechen Gao, BOQI ZHAO, Andrew Lee, Ian Chuang, Hanchu Zhou, Hang Wang, Zhe Zhao, Junshan Zhang, Iman Soltani
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 flow latent decoding (FLD) (ablation) | VITA with action AE · ThreadNeedle | success rate (%): 0 -> 92 | Improved |
| replace zero-padding with learned action autoencoder latents (ablation) | VITA · ThreadNeedle | success rate (%): 0 -> 92 | Improved |
| freeze pre-trained action autoencoder (ablation) | VITA · online success and offline MSE | success rate | Hurt |
| add FLC alongside FLD (ablation) | VITA · simulated tasks | success rate | Improved |
| use FLC alone instead of FLD (ablation) | VITA · simulated tasks | success rate | Hurt |
| add contrastive vision-action loss (ablation) | VITA · some tasks | success rate | Mixed |
| replace action AE with variational AE (KL term) (ablation) | VITA · ThreadNeedle | success rate | Hurt |
| make vision encoder variational (ablation) | VITA · ThreadNeedle | success rate | Hurt |
| use VITA instead of FM transformer with AdaLN | FM · inference latency, vector latents | latency (ms/chunk): 0.3307 -> 0.2215 | Improved |
| use VITA instead of FM with cross-attention | FM · inference latency, grid latents | latency (ms/chunk): 0.5102 -> 0.2502 | Improved |
| use MLP-only FM instead of VITA (ablation) | FM · PushT | success rate (%): 88 -> 0 | Hurt |
| use VITA instead of FM | FM · SlotInsertion | success rate (%): 82 -> 78 | Hurt |
Closest papers by meaning
- Reflex: Real-Time Vision-Language-Action Control through Streaming Inference (ICML 2026 · similarity 0.90)
- LAFP: Preserving Latent Action Structure in Latent Policy Learning via Flow Matching (NeurIPS 2026 · similarity 0.88)
- Contextual Flow Matching: Adaptive Step Selection in Flow Models for Efficient Visual Generation (NeurIPS 2026 · similarity 0.87)
- Flow-GRPO: Training Flow Matching Models via Online RL (NeurIPS 2025 · similarity 0.86)
- FlowPolicy: Enabling Fast and Robust 3D Flow-Based Policy via Consistency Flow Matching for Robot Manipulation (AAAI 2025 · similarity 0.86)
- RotVLA: Rotational Latent Action for Vision-Language-Action Model (NeurIPS 2026 · similarity 0.86)
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.
