A BP Network Control Approach for QoS-Aware MAC in Cloud Robotics

Abstract

The basic idea of Cloud Robotics is dynamically uploading the compute-intensive applications to the cloud, which greatly enhances the intelligence of robots for the high processing and parallel ability of cloud. However, for the nature of uncertainty of mobility, different kinds of applications on robot may have different Quality of service (QoS). The paper proposes a BP network for QoS-aware MAC(BPFD-MAC) in Cloud Robotics form a view control theory, which can support both absolute and relative QoS guarantees while the energy saving. The hard and soft QoS constraints are de-coupled by normalized into a two-level cascade feedback loop. The former is Active Time Loop (AT-Loop) to enforce the absolute QoS guarantee for real-time application and the later is Contention Window Loop (CW-Loop) to enforce the relative QoS guarantee for Best Effort traffics. Finally, the Back-propagating (BP) neuron network based PID is used for self-tuning parameters and controller design. The hardware experiments demonstrate the feasibility of BPFD-MAC. Comparing with FD-MAC, BPFD-MAC has new feature of absolute QoS support and further developed two advantages:In the condition of heavy loads, BPFD have about 18% great throughput and 14% great power efficient; and in light load, BPFD have lower total energy consumption.