Reinforcement Learning-Aided Edge Intelligence Framework for Delay-Sensitive Industrial Applications

Abstract

With the advancement in next-generation communication technologies, the so-called Tactile Internet is getting more attention due to its smart applications, such as haptic-enabled teleoperation systems. The stringent requirements such as delay, jitter, and packet loss of these delay-sensitive and loss-intolerant applications make it more challenging to ensure the Quality of Service (QoS) and Quality of Experience (QoE). In this regard, different haptic codec and control schemes were proposed for QoS and QoE provisioning in the Tactile Internet. However, they maximize the QoE while degrading the system’s stability under varying delays and high packet rates. In this paper, we present a reinforcement learning-based Intelligent Tactile Edge (ITE) framework to ensure both transparency and stability of teleoperation systems with high packet rates and variable time delay communication networks. The proposed ITE first estimates the network challenges, including communication delay, jitter, and packet loss, and then utilizes a Q-learning algorithm to select the optimal haptic codec scheme to reduce network load. The proposed framework aims to explore the optimal relationship between QoS and QoE parameters and make the tradeoff between stability and transparency during teleoperations. The simulation result indicates that the proposed strategy chooses the optimal scheme under different network impairments corresponding to the congestion level in the communication network while improving the QoS and maximizing the QoE. The end-to-end performance of throughput (1.5 Mbps) and average RTT (70 ms) during haptic communication is achieved with a learning rate and discounted factor value of 0.5 and 0.8, respectively. The results indicate that the communication system can successfully achieve the QoS and QoE requirements by employing the proposed ITE framework.