Scaling Unlabeled Video-Audio Pretraining for Few-Shot Latent Action Models

This report synthesises findings from 5 peer-reviewed papers addressing the following research question: How does the scaling of unlabeled video-audio pretraining data affect the few-shot adaptation accuracy of latent action models on the RoboBench benchmark compared to supervised baselines. 5 claims were extracted from source literature; 5 were independently verified against retrieved documents. An automated multi-reviewer quality assessment produced a score of 9.0/10. This report is a machine-generated literature synthesis and does not constitute original research.

Research goal: How does the scaling of unlabeled video-audio pretraining data affect the few-shot adaptation accuracy of latent action models on the RoboBench benchmark compared to supervised baselines?

Autonomous literature synthesis. Automated review score: 9.0/10. Full text and citation available at Assignee Research.