EgoWalk: A Multimodal Dataset for Robot Navigation in the Wild
🤖 AI Summary
Existing public navigation datasets lack human navigation behavior data covering indoor/outdoor environments, multiple seasons, and diverse geographic regions, limiting the generalizability of end-to-end models. To address this, EgoWalk introduces a 50-hour real-world, human-robot collaborative navigation dataset, the first to synchronously capture multimodal sensor data (RGB-D, IMU, GPS). It proposes an automated annotation pipeline that generates natural-language goal instructions and traversability segmentation masks. The dataset is systematically quantified for diversity, and a novel benchmarking protocol is established—specifically designed for realistic navigation tasks. All data, processing code, and hardware documentation are fully open-sourced. Experiments demonstrate that EgoWalk significantly outperforms existing datasets in cross-environment generalization and downstream tasks—including instruction-following navigation and semantic map construction—thereby advancing robust, real-world navigation research.
Application Category
📝 Abstract
Data-driven navigation algorithms are critically dependent on large-scale, high-quality real-world data collection for successful training and robust performance in realistic and uncontrolled conditions. To enhance the growing family of navigation-related real-world datasets, we introduce EgoWalk - a dataset of 50 hours of human navigation in a diverse set of indoor/outdoor, varied seasons, and location environments. Along with the raw and Imitation Learning-ready data, we introduce several pipelines to automatically create subsidiary datasets for other navigation-related tasks, namely natural language goal annotations and traversability segmentation masks. Diversity studies, use cases, and benchmarks for the proposed dataset are provided to demonstrate its practical applicability. We openly release all data processing pipelines and the description of the hardware platform used for data collection to support future research and development in robot navigation systems.
Problem
Research questions and friction points this paper is trying to address.
Lack of large-scale real-world navigation datasets for training robots
Need for diverse environments in robot navigation data collection
Absence of automated pipelines for navigation-related task datasets
Innovation
Methods, ideas, or system contributions that make the work stand out.
Multimodal dataset for diverse navigation environments
Automated pipelines for subsidiary navigation datasets
Open release of data processing and hardware details
🔎 Similar Papers
No similar papers found.
