Reinforcement Learning Based Backoff Management for Industry 5.0

Industry 5.0 marks a transition from the digitalization focus of Industry 4.0 to a paradigm emphasizing resilience, sustainability, and human-centric processes. In such dynamic networks, reinforcement learning (RL) algorithms can play a crucial role in enhancing performance. The paper proposes a novel approach using a centralized RL algorithm to optimize the medium access control for a moving autonomous guided vehicle (AGV) on an industrial shop floor. This ensures uninterrupted production flow by dynamically managing the network traffic. The use case scenario considers a mobile AGV transmitting data to a base station (BS) within a harsh industrial environment. It uses the RL algorithm to dynamically select an optimal backoff (BO) time for an ALOHA-like channel access protocol. This enables accurate data transmission without prior knowledge of the industrial environment. The results show an improvement of up to 34.6% in success probability compared to traditional BO design approaches. The RL model achieves outstanding performance, guaranteeing a minimum success probability of 99.46%.