Robotics · 2026-10-14
$\Delta$ Flow Matching for Robot Foundation Models
Kuangji Zuo, Tuo An, Xinying Guo, Gen Li, Mengfei Zhao, Jiaqi Bai, Bofan Lyu, Zhuoyuan Yu, Jinghan Yang, Jiayi Li, Jindou Jia, Jianfei Yang
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
The authors propose Δ-flow matching for robot foundation models, replacing the fixed Gaussian source used in flow matching with a source constructed from recent history plus mild noise. This reduces transport to a residual Δ, enables single-step inference, improves training efficiency and generalization on LIBERO/LIBERO-Plus, preserves action smoothness, and can adapt pretrained WAMs using action-history sources without retraining.
Why it matters
Robot foundation models using flow matching must learn and sample transports from Gaussian noise. By using history-based sources and learning residual changes instead of full transports, Δ-flow targets both learning cost and deployment latency. The paper reports faster convergence/epochs and single-step inference speedups while maintaining or improving task success, plus smoother actions measured via reduced action jerk.
Method
- Construct flow sources from recent executed history (actions and, for WAMs, visual context) with mild Gaussian perturbation, instead of a fixed Gaussian prior, to learn residual changes Δ.
- Supervise one-step generation directly by applying one Euler update from the history-based source at τ=1 to τ=0 with endpoint consistency loss.
- Integrate Δ-flow into VLA π0.5 and WAM policies (ImageWAM, Fast-WAM), producing future visual representations and actions in a single flow step at inference (with one-step Euler update).
Limitation
The paper does not state a limitation section in the provided text.
Abstract (from arXiv)
Flow matching is a common generative recipe for robot foundation models. Vision-language-action models (VLAs) generate action chunks, while world-action models (WAMs) additionally generate future visual states. Both start generation from a fixed Gaussian prior, despite temporal continuity between past and future states and actions. This requires costly training and iterative sampling, especially for WAMs with an additional visual branch. We introduce $\Delta$-flow, which constructs the source from recent history with mild noise rather than from a fixed Gaussian prior. This moves the source closer to the target, reducing transport to a small residual $\Delta$ that can be learned efficiently and enables single-step inference. We integrate $\Delta$-flow into the VLA policy $\pi_{0.5}$ and the WAM policies ImageWAM and Fast-WAM. On LIBERO, ImageWAM reaches 90% success in one epoch rather than six, and Fast-WAM reaches 95% in two epochs rather than eight. With LoRA fine-tuning, $\pi_{0.5}$ exceeds its baseline's five-epoch performance after one epoch. Single-step inference gives $1.54\times$, $2.49\times$, and $4.45\times$ speedups on $\pi_{0.5}$, ImageWAM, and Fast-WAM, respectively. It preserves action smoothness, achieving a $4.5\times$ reduction in action jerk compared with one-step Gaussian sampling. $\Delta$-flow reaches state-of-the-art generalization performance on LIBERO-Plus. Action-history sources can be applied to pretrained WAM policies without retraining. Replacing the Gaussian action source at inference largely preserves task success with a single sampling step and can slightly improve it. We further study how action representation and visual information in action learning affect history-to-future prediction in $\Delta$-flow. These results show that $\Delta$-flow is simple yet effective, improving the efficiency and generalization of robot foundation models.
Related papers
- {\Delta}WAM: Distilling Action Tangent Fields into World Action Models
- PRIME W0: Physical Priors from World Models for Embodied Manipulation
- Vision-Language-Delta-Action: Learning Compact Deltas as Actionable Transition Interfaces for Robotic Manipulation
- Native Action-Prior Learning from Videos for World Action Models
- EvoMem-VLA: State-Evolution Memory for Long-Horizon Robot Manipulation
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-11 · Indexed by Pub-AI: 2026-10-14 UTC
$Δ$ Flow Matching for Robot Foundation Models
Preprint posted: 2026-10-11 · Original record ↗
- arxiv
- 2610.15794
Explicit source versions (1)
- v1 · 2026-10-11
- Preprint available
Source date: 2026-10-11 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Flow matching is a common generative recipe for robot foundation models. Vision-language-action models (VLAs) generate action chunks, while world-action models (WAMs) additionally generate future visual states. Both start generation from a fixed Gaussian prior, despite temporal continuity between past and future states and actions. This requires costly training and iterative sampling, especially for WAMs with an additional visual branch. We introduce $Δ$-flow, which constructs the source from recent history with mild noise rather than from a fixed Gaussian prior. This moves the source closer to the target, reducing transport to a small residual $Δ$ that can be learned efficiently and enables single-step inference. We integrate $Δ$-flow into the VLA policy $π_{0.5}$ and the WAM policies ImageWAM and Fast-WAM. On LIBERO, ImageWAM reaches 90% success in one epoch rather than six, and Fast-WAM reaches 95% in two epochs rather than eight. With LoRA fine-tuning, $π_{0.5}$ exceeds its baseline's five-epoch performance after one epoch. Single-step inference gives $1.54\times$, $2.49\times$, and $4.45\times$ speedups on $π_{0.5}$, ImageWAM, and Fast-WAM, respectively. It preserves action smoothness, achieving a $4.5\times$ reduction in action jerk compared with one-step Gaussian sampling. $Δ$-flow reaches state-of-the-art generalization performance on LIBERO-Plus. Action-history sources can be applied to pretrained WAM policies without retraining. Replacing the Gaussian action source at inference largely preserves task success with a single sampling step and can slightly improve it. We further study how action representation and visual information in action learning affect history-to-future prediction in $Δ$-flow. These results show that $Δ$-flow is simple yet effective, improving the efficiency and generalization of robot foundation models.
