Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Curating Demonstrations using Online Experience

Created by
  • Haebom

Author

Annie S. Chen, Alec M. Lessing, Yuejiang Liu, Chelsea Finn

Outline

In this paper, we propose Demo-SCORE, a novel method for self-healing demos based on online robot experiences, to address the heterogeneity problem of robot demo datasets containing demo datasets with various qualities and levels. Demo-SCORE learns a classifier that distinguishes successful policy executions from failed executions and cross-validates it to filter out heterogeneous demo datasets. Experimental results show that Demo-SCORE effectively identifies inefficient demos without manual cleaning, and improves the success rate by 15-35% compared to policies trained with all existing demos.

Takeaways, Limitations

•
Takeaways:
◦
We present a novel method to effectively address the challenges of heterogeneous robot demo datasets.
◦
Save time and money by manually refining your demos.
◦
Significantly improves the success rate of robot policies (15-35% or more).
◦
Presenting the possibility of building an online experience-based self-healing system.
•
Limitations:
◦
Further research is needed on the generalization performance of the proposed method.
◦
Applicability evaluation for various tasks and robotic platforms is needed.
◦
The performance of the classifier is highly dependent on the performance of Demo-SCORE.
◦
The difficulty of collecting robot experience data in real-world environments.
👍
Made with Slashpage