BTreeFuzz: Enhanced Feedback Mechanism for ROS Program Fuzzer Based on Behavior Tree

Robots are increasingly taking on human roles and functions, and societal expectations for these machines to perform critical tasks and eliminate the problem of human error continue to rise. The robot operating system (ROS) is considered the de facto standard that has been widely used to develop robotic programs. However, even though robot malfunctions should be minimized, it remains highly challenging to develop ROS programs without software vulnerabilities. As a result, many ROS programs still contain software vulnerabilities that can lead to serious malfunctions. To address this concern, researchers have recently presented several fuzzing techniques to automatically detect these vulnerabilities in ROS programs. A feedback mechanism for fuzzing requires a specific metric tailored to the features of ROS programs. However, existing ROS program fuzzers have been unable to provide adequate feedback mechanisms to make this possible. For instance, ROFER, a state-of-the-art ROS program fuzzer, cannot correctly differentiate between multiple distinct robot statuses based on its feedback mechanism metric, and often mistakenly identifies identical statuses as being different from each other, making it ineffective at detecting unique crashes.