Efficient Resource Augmentation of Resource Constrained UAVs Through EdgeCPS

Abstract

We propose an efficient Resource Augmentation Framework (RAF) for resource-constrained UAVs through EdgeCPS. Typical UAVs with small form factors have limited computation power which hinders their ability to perform critical or computation-intensive missions. By exploiting EdgeCPS, UAVs can get computational support from the EdgeCPS and diversity its missions. Existing solutions allow exploiting the EdgeCPS; however, the network overhead is too great that it cannot be adopted in resource-constrained UAVs. The proposed framework, RAF, provides Task Management Module (TMM) and Offloading Inference Module (OIM) to solve the issue. Using Raspberry Pi 4 as the mission computer for the UAV, RAF shows an inference performance of 11.9 FPS in the ResNet-18 model, whereas the existing work shows about 6 FPS.