$Δ$ Flow Matching for Robot Foundation Models
Flow matching is a common generative recipe for robot foundation models.
ProofPaper ↗
Key points
- Vision-language-action models (VLAs) generate action chunks, while world-action models (WAMs) additionally generate future visual states.
- We introduce $Δ$-flow, which constructs the source from recent history with mild noise rather than from a fixed Gaussian prior.
- We integrate $Δ$-flow into the VLA policy $π{0.5}$ and the WAM policies ImageWAM and Fast-WAM.
- We further study how action representation and visual information in action learning affect history-to-future prediction in $Δ$-flow.
Sources (1)
- [1]$Δ$ Flow Matching for Robot Foundation ModelsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 11, 05:31 PM
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.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 11, 2026A Persian Vision-Language-Action Dataset and Benchmark for Ambiguity- and Infeasibility-Aware Instruction-to-Plan Grounding
- Oct 11, 2026StreamingWAM: Streaming World Action Modeling via Chunk-Wise Staircase Flow Matching
- Oct 10, 2026Verify, Act, Distill: Turning Test-Time Compute into Test-Time Adaptation for Vision-Language-Action Models
- Oct 10, 2026BridgeWAM: Bridging Video and Action Experts via Latent Bridge Queries
- Oct 9, 2026Dream to Remember: Continual Steering of Generalist Robot Policies with World Models
- Oct 9, 2026Rethinking Relative Policy Differences Transfer of On-Policy Distillation