Generative Models for Temporal-Based Task Definition

Autonomous systems rely on artificial intelligence to perform their tasks more effectively. With the increasing complexity of tasks, it is essential to provide a structured way to define tasks. This paper explores a novel approach to extract task definitions with temporal information using generative models. The approach combines the strengths of the T5 and BERT models to understand and sequence tasks. Specifically, the BERT model classifies the temporal relationship between tasks before the T5 model is used to generate the DSL sequence. Furthermore, we present a simple DSL for a household robot use-case and a dataset of natural language commands with temporal information. The model is trained on this dataset and evaluated on its ability to generate accurate and temporally ordered task definitions.