A Convolutional Neural Network for Goal Recognition

Goal recognition is a complex process that involves deciphering objectives from observable cues, allowing for a deeper understanding of human or machine intentions. It revolves around the fundamental concept of goal inference, focusing not on the specific actions themselves but rather on the underlying purpose or aim behind those actions. Many approaches to goal recognition typically require careful domain design of the agent’s potential actions within an environment by a domain expert or involve costly real-time computations, or sometimes both. Recently, alternative goal recognition approaches have emerged with the aim of eliminating the need for meticulous, manual domain design and the reliance on expensive online planners executions. One of these approaches to model-free goal recognition involves machine learning classification tasks, which is the approach we have adopted in our work. The purpose of our work is to present a straightforward Convolutional Neural Network based architecture that achieves better goal-prediction accuracy and timeliness, as demonstrated through rigorous experimental analysis on standard evaluation domains.