Dihong Huang

Quality-aware pretraining strategies, particularly approach-aware episode conditioning, can significantly outperform plain behavioral cloning in VLA tasks, with optimal choices depending on label granularity and model architecture.

Continual pretraining on AXIS boosts robot manipulation success rates by 5.8% over previous methods and outperforms existing datasets by over 37%.

Substantial challenges in task generalization and visuomotor robustness are revealed, highlighting critical gaps in current dexterous manipulation benchmarks.