EgoSteer Enables Steerable Dexterous Manipulation via Egocentric Video
June 20, 2026
EgoSteer scales dexterous Vision-Language-Action (VLA) pre-training using the EgoSmith pipeline, which converts 9.6K hours of egocentric video into high-quality training data. The system achieves 9x higher throughput than prior state-of-the-art methods to enable language-guided dexterous manipulation.
HOW THIS AFFECTS YOU
●
builderYou can leverage higher-throughput data pipelines for training robot manipulation policies.
●
researcherThe integration of world models with VLAs offers a new pathway for dexterous robot scaling.