RoboPapers Ep#17 EgoZero: Robot Learning from Smart Glasses
Jun 26, 2025
Join Vincent Liu, a Stanford math and computer science graduate, and Ademi Adeniji, a UC Berkeley PhD student, as they dive into their groundbreaking work on robot learning through smart glasses. They discuss the vital role of human interaction in enhancing robot intelligence and the challenges of mapping movements to robotic actions. The duo also tackles the costs and benefits of advanced smart glasses, along with the complexities of using third-party data for accurate robotic learning. Tune in for insights into robotic evolution and future tech innovations!
AI Snips
Chapters
Transcript
Episode notes
Limitations of Teleoperation and Data Sources
- Teleoperation is inefficient for diverse, large-scale robot data collection due to required training and lack of ergonomic embodiment.
- Human-in-the-wild egocentric data from smart glasses offers a scalable, natural source of robot training data.
Unified 3D Representation Benefits
- Using a unified 3D point representation allows encoding human and robot morphologies in a common space.
- This morphology-agnostic 3D to 3D mapping simplifies policy learning and enhances data augmentation.
Use 3D Augmentation for Generalization
- Augment 3D point demonstrations with random transformations to improve generalization at inference.
- Ensure inference data points fall within the augmented training distribution to maintain success.