Adaptive Sampling in Cloud-Based Cyber-Physical Systems (CBCPS) Through Reinforcement Learning

Cloud-Based Cyber-Physical Systems (CBCPS) offer considerable benefits by harnessing computational and storage capacities in the cloud. However, disruptions in obtaining real- time access to cloud computational and network resources can impede the control of CBCPS. Hence, minimizing resource demands by adapting the controller's sampling rate can enhance performance. Due to the dynamic nature of the system model, we employ reinforcement learning techniques. We demonstrate our approach through simulation to remotely control a car's speed using cloud-based controllers, dynamically adjusting the sampling frequency to reduce network load and optimize cloud resource utilization. Our proposed method utilizes an actor-critic reinforcement learning framework, specifically the Deterministic Policy Gradient (DDPG) algorithm, to learn optimal policies that strike a balance between resource usage, response time, real-time performance, and system stability.