EgoBridge: Domain Adaptation for Generalizable Imitation from Egocentric Human Data

2025-09-23

AI Summary

To bridge the visual, sensor modality, and kinematic discrepancies between human egocentric videos and robotic execution, this paper introduces a novel paradigm—cross-embodiment imitation learning. Our core contribution is EgoBridge, a unified co-training framework that jointly optimizes visual representation learning, policy mapping, and kinematic decoupling via domain adaptation. Crucially, EgoBridge employs optimal transport to align latent policy spaces across human and robot embodiments, preserving action semantics while minimizing domain shift. Evaluated on three real-world manipulation tasks, our method achieves an absolute 44% improvement in policy success rate. More significantly, it demonstrates strong generalization to unseen objects, scenes, and tasks—whereas existing baselines fail completely. This work provides the first scalable and interpretable solution for robust knowledge transfer from first-person human demonstrations to robotic policies.

Technology Category

• Machine Learning: Imitation Learning & Inverse Reinforcement Learning
• Intelligent Robots: Embodied AI
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: The sharing economy
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Egocentric human experience data presents a vast resource for scaling up end-to-end imitation learning for robotic manipulation. However, significant domain gaps in visual appearance, sensor modalities, and kinematics between human and robot impede knowledge transfer. This paper presents EgoBridge, a unified co-training framework that explicitly aligns the policy latent spaces between human and robot data using domain adaptation. Through a measure of discrepancy on the joint policy latent features and actions based on Optimal Transport (OT), we learn observation representations that not only align between the human and robot domain but also preserve the action-relevant information critical for policy learning. EgoBridge achieves a significant absolute policy success rate improvement by 44% over human-augmented cross-embodiment baselines in three real-world single-arm and bimanual manipulation tasks. EgoBridge also generalizes to new objects, scenes, and tasks seen only in human data, where baselines fail entirely. Videos and additional information can be found at https://ego-bridge.github.io

Problem

• Bridging domain gaps between human and robot data for imitation learning
• Aligning policy latent spaces across different visual and kinematic domains
• Enabling robots to generalize manipulation skills from egocentric human data

Innovation

• Domain adaptation aligns human-robot policy latent spaces
• Optimal Transport measures feature-action discrepancy for alignment
• Co-training framework generalizes to unseen objects and scenes