Agentic data science with the Data Agent Kit
cloud.google.com·7h ago
AI Research
arXiv CS.AI·Jul 27·5 min read
ACME: A Multi-Cultural, Multi-Embodiment Social-Navigation Dataset
TL;DR
ACME is a new dataset designed to enhance social navigation research by capturing diverse human-robot interactions across different cultures and environments. It includes 29.35 hours of robot data and 43.5 hours of pedestrian tracking, focusing on complex social scenarios.
✦ Why It Matters
Utilize the ACME dataset to improve your robot's social navigation capabilities in diverse environments.
Key Takeaways
- Explore the ACME dataset for diverse social navigation scenarios.
- Incorporate robot speech in your navigation policies.
- Use 3D and 2D scene features for enhanced robot perception.
- Analyze pedestrian behaviors to improve interaction strategies.
Full Summary
ACME, which stands for Cross-cultural, Multi-Embodiment dataset, addresses the lack of diversity in existing datasets for social navigation by collecting data from 8 sites in 5 countries using 7 different robot embodiments. This large-scale effort resulted in 29.35 hours of onboard robot data and 43.5 hours of overhead pedestrian tracking data.
The dataset emphasizes goal-driven social navigation behavior, particularly in scenarios involving explicit robot-crowd interactions, including robot speech. ACME provides rich features such as 3D and 2D scene data, odometry, interaction information, and human-annotated pedestrian trajectory labels.
Its qualitative and quantitative analyses demonstrate that ACME captures more challenging scenarios and a wider range of pedestrian behaviors compared to previous datasets, making it a valuable resource for developing navigation policies and predicting pedestrian movements.