Message passing-based inference in an autoregressive active inference agent

Abstract

We present the design of an autoregressive active inference agent in the form of message passing on a factor graph. Expected free energy is derived and distributed across a planning graph. The proposed agent is validated on a robot navigation task, demonstrating exploration and exploitation in a continuous-valued observation space with bounded continuous-valued actions. Compared to a classical optimal controller, the agent modulates action based on predictive uncertainty, arriving later but with a better model of the robot's dynamics.

14 pages, 4 figures, proceedings of the International Workshop on Active Inference 2025. Erratum v1: in Eq. (50), $p(y_t,\hat{I}˜,u_t\mid y_{\ast },{\mathcal{D}}_k)$ should have been $p(y_t,\hat{I}˜\mid u_t,y_{\ast },{\mathcal{D}}_k)$