Low-cost fuzzing drone control system for configuration errors threatening flight safety in edge terminals

With the wide deployment of edge-cloud computing, drones have drawn much attention to serving as edge nodes responsible for data collection. Numerous configuration parameters play a crucial role in regulating the attitude and altitude of the drone. If these parameters are misconfigured, drones will fall into abnormal flight states, such as trajectory deviation, and even crash to the ground. Previous research primarily addresses system memory errors which result in obvious system failures, but does not effectively detect the anomalies in drone flight states. Moreover, a large amount of computational cost on the detection will significantly increase the power consumption of drones, which violates the principle of green computing. This paper focuses on abnormal drone flight states caused by configuration parameter errors. We propose a novel state-guided fuzzing system called LDFuzzer, which searches for incorrect configuration parameter values that would trigger abnormal flight states. To enhance the capability of searching for multiple optimal solutions, we design a Quality-Diversity-Enhanced Genetic Algorithm (QDGA) to mutate configuration values. Moreover, we also propose a set of new test oracles to detect abnormal flight states of drones in real time with restricted computational resources. We evaluated LDFuzzer on the drone control system ArduPilot and successfully discovered 3399 incorrect configuration parameter values. Additionally, the results from our experiments show that LDFuzzer can automatically analyze fuzzing outcomes and has identified 8 software bugs linked to configuration parameters.