CrossEdgeIM: An Edge-Based Approach for Interactive Robotic Behavior Model Discovery

With the rapid advancement of the Internet of Robotic Things (IoRT), interactive robotic systems are increasingly deployed in distributed edge environments to support real-time human–robot collaboration and adaptive task execution. However, conventional robotic behavior modeling approaches typically rely on centralized data collection, which introduces latency, high communication overhead, and privacy risks. To address these challenges, this paper proposes CrossEdgeIM, an edge-based approach for discovering interactive robotic behavior models. In CrossEdgeIM, robotic activity nodes at the edge continuously capture and preprocess event features during interactions, while organizational nodes aggregate and transform these features into local robotic behavior models. A central node then integrates the local models into a unified global behavior model through structured model merging. This design significantly reduces communication costs, enhances privacy protection, and improves the accuracy of behavior discovery in IoRT scenarios. Experimental results demonstrate that CrossEdgeIM achieves substantial gains in both interpretability and scalability compared to conventional centralized methods, offering an effective solution for Edge AI–driven robotic behavior discovery in distributed environments.