Learning Transferable World-Action Models from Task-Paired Human Videos
Overview
WATCH, a cross-video pre-training method, improves robot performance using human demonstrations. It outperforms baselines, boosting MimicDroid success to 77.7% with Cosmos3. Results show cross-video prediction enhances transferable world-action models without needing human-robot pairs. arXiv:2610.14351v1 Latest: WATCH improves MimicDroid success to 77.7% with Cosmos3, showing cross-video prediction's effectiveness for transferable world-action models.
Generated from attributed reports · Updated 46 minutes ago
Event evidence and corrections
0 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
No current evidence-backed claims. Missing information remains not reported.
Report timeline
Follow attributed reports and material updates.
- arXiv Robotics — research abstractsLearning Transferable World-Action Models from Task-Paired Human Videos
The paper presents WATCH, a cross-video pre-training method that learns transferable world-action models from task-paired human demonstrations. By jointly predicting actions and future frames across different demonstrations, WATCH achieves improved motion alignment and better performance on unseen robots, outperforming within-video pre-training and co-training baselines.
Event coverage history
There is not enough continuous observation data to show a trend.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).