Expected Free Energy-based Planning as Variational Inference

The paper presents a novel formulation of Expected Free Energy (EFE) planning as Variational Free Energy minimization, integrating epistemic priors into a generative model. Key findings include a decomposition of the Variational Free Energy functional into expected plan costs and a complexity term, reinforcing alignment with the Free Energy Principle. The approach was validated across three environments, showing that it enhances information-seeking behavior and outperforms traditional plan-based methods in stochastic scenarios, suggesting improved scalability for practical applications in uncertain environments.