Theory of mind is an essential ability for complex social interaction and collaboration. Researchers in cognitive science and psychology have previously sought to integrate theory of mind capabilities into artificial intelligence (AI) agents to improve collaborative abilities (Cuzzolin, Morelli, Cirstea, & Sahakian, 2020). These approaches, however, are hampered by the need for labor-intensive hand-labeling of datasets, which prevents them from scaling up to large, real-world datasets. To address this challenge, we introduce the Recurrent Conditional Variational Autoencoder (RCVAE), a novel model designed to predict intent from human behavioral trajectories without the prerequisite of hand-labeled data. We show that in the Overcooked-AI environment, the RCVAE outperforms baseline Long Short-Term Memory (LSTM) models in predicting intent, achieving higher prediction accuracy and greater predictive stability. The implications of these results are significant; the RCVAE's proficiency in learning the relationship between basic actions and resulting contextual behaviors, without needing hand-labeled data, will be crucial for scaling from simple to complex, real-world environments.

Authors:

Willa Mannering: Johns Hopkins Applied Physics Laboratory; Noah Ford: Johns Hopkins University Applied Physics Lab; Justin J Harsono: Johns Hopkins University Applied Physics Laboratory; John Winder: Johns Hopkins University Applied Physics Laboratory