Multi-Label Behavior Cloning for Full-Area Exploration in IoT-Enabled Unstructured Unknown Environments

The widespread application of Internet of Things (IoT) technology provides connectivity and intelligence in mo-bile terminals, which supports their independent exploration in unknown environments. This paper focuses on the full-area exploration in unstructured unknown environment. To tackle the complexities and dynamic nature of unknown environments, we propose a full-area exploration method based on multi-label behavior cloning (MBC). The MBC innovatively uses behavior-related rewards as weights for labels to enhance the ability of the model to learn effective behaviors. We develop a lightweight neural network model that employs multiple attention mechanisms to fit the mapping from state to action in the behavior cloning process. The lightweight model significantly reduces the computational load, making it suitable for deployment on terminal devices with limited computational resources. Additionally, we refine input handling in the model by introducing interpolation of perceived states, which standardizes the dimensions of input data. Finally, we demonstrate the effectiveness of our proposed exploration method through experiments.