A Framework for Dynamic Configuration of Cellular Industrial Networks in Mobile Robotics Applications

Digitalization of industries is essential for achieving the ever-increasing demands for flexible and customized production. Wireless communication plays a key role in enabling digitalization of industries. Cellular communication systems are expected to efficiently support data traffic with varying sets of requirements on latency, reliability and throughput as well as mobility and scalability in industrial applications. In this paper, we describe a framework that enables closed-loop runtime adaptation of a cellular network configuration using artificial intelligence and machine learning ($\mathrm{AI} / \mathrm{ML}$) methods based on the observed network performance characteristics and dynamic changes in application service level requirements. While most of the existing works on the use of $\mathrm{AI} / \mathrm{ML}$ methods for 5 G performance enhancement are based on computer simulations, our framework has been experimentally evaluated on an industrial 5G non-public network in an industrial environment. We present over-the-air empirical performance results of a 5 G non-public network deployment in the local industrial spectrum in Germany. Our empirical results give insights into 5G deployments in industrial environments and highlight that the desired communication robustness for a mobile robotics use case can be achieved with efficient use of radio resources using deep reinforcement learning.