A Novel Collaborative Reinforcement Learning Method for Human-to-Avatar Interactions

The rapid development of television technology has led to the development of immersive human-avatar interaction systems. This study focuses on enabling service migration and resource allocation strategies necessary for this network. The telepresence suite enables users to interact with virtual environments and avatars in real time, requiring tactile feedback, motion capture, and improvements in AI-driven gesture recognition. This paper presents a comparison between two service migration algorithms: the round-robin (RR) method and the new collaborative reinforcement learning (ML-MARL) method. While RR provides simplicity, ML-Marl exhibits excellent flexibility and performance by intelligently allocating resources in dynamic situations. Experimental results show that the use of ML-MARl reduces the latency by about 65% compared to RR, highlighting potential applications in telemedicine, education, manufacturing, and military applications. The findings highlight the importance of adaptive learning strategies for optimizing next-generation remote control systems. Paving the method of deliberative virtual interactions highly responsive in environments requiring seamless connectivity and real-time data. © 2025 Elsevier B.V., All rights reserved.