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.

Proximal Control of UAVs with Federated Learning for Human-Robot Collaborative Domains

Created by
  • Haebom

Author

Lucas Nogueira Nobrega, Ewerton de Oliveira, Martin Saska, Tiago Nascimento

Outline

In this paper, we propose an action recognition and control method that uses two long short-term memory (LSTM) deep neural networks and three fully-connected layers to solve complex command (action) classification problems in human-robot interaction (HRI). In particular, we apply federated learning (FL) to enable distributed learning in a multi-robot scenario consisting of multiple drones, thereby performing learning without access to the cloud or other storage. The experimental results using real robots show that the accuracy is over 96%, which solves the problem of human operator occlusion that may occur when operating drones.

Takeaways, Limitations

•
Takeaways:
◦
The learning efficiency of a multi-robot system was improved through distributed learning based on federated learning.
◦
It effectively solves the occlusion problem that occurs when operating a drone, thereby increasing the possibility of practical application.
◦
It achieved high accuracy (over 96%), demonstrating its applicability to actual HRI systems.
•
Limitations:
◦
Further evaluation of the generalization performance of the proposed method is needed.
◦
Further research is needed on robustness in various environments and conditions.
◦
There is a lack of detailed description of the LSTM architecture used and parameter optimization of federated learning.
◦
Performance evaluation for various types of instructions is required.
👍
Made with Slashpage