SWAM studies how action learning affects language-video alignment
Authors report 88.10% easy and 61.56% hard RoboTwin success, and 40.00% versus 23.33% on three real tasks.
48
Product
RoboTwin
- Mentions · 7d1
- Mentions · 30d3
- All mentions8
- First seenSep 18, 2026
- Last mentioned3d ago
Timeline
8 articles · newest firstJust Action Models match world-action policies without visual generation
Against FastWAM, reported LIBERO success moves from 97.0%, 55.3%, and 91.2% to 98.2%, 64.5%, and 91.4%.
46
VLA-Δ: Learning Physical State Transition for Vision-Language-Action Models
VLA-Δ trains robot policies on physical state transitions and beats π0.5 in real-robot tests.
38
RoboRSI: Stable, efficient, and reusable robot self-evolution in complex real-world environments
RoboRSI lets a mobile manipulator reuse validated skills, beating baselines by up to 11 points.
38
TAPDreamer: Transferable Adversarial Patches for World Action Models
TAPDreamer crafts transferable adversarial patches that drop world-action robot success to near zero without querying policies.
62
WorldLine: Action-Driven Visual Simulation for Robotic Manipulation
WorldLine predicts robot action outcomes from 10,000-plus hours of video across many embodiments.
52
Rolling-WAM: World Action Models with Rolling Imagination
Rolling-WAM speeds robot replanning 4.5x by staggering joint video-action denoising across cycles.
44
Transferring the Intelligence of VLMs to Robotic Control
RoboDawn lets a VLM control robots via discrete commands, beating trained policies zero-shot and one-shot.
52
Appears with
- LIBEROproduct · 5
- LIBERO-Plusproduct · 2
- pi0.5ai model · 2
- FastWAMai model · 2
- ImageWAMai model · 1
- VLA-Deltaai model · 1
- RoboDawnproduct · 1
- Frankaproduct · 1
- Motusai model · 1
- ZJU-OmniAIorganization · 1
- RoboDojoproduct · 1
- Just Action Modelai model · 1
- SWAMai model · 1
- LIBERO-PROproduct · 1
- DreamWAMai model · 1
- Unitree G1product · 1
Entities are extracted by the model from each article. First seen , last mentioned . Watching keeps it in this browser only (no account); the watchlist page and dashboard alerts use it.