Message passing-based inference in an autoregressive active inference agent

2025-09-29

AI Summary

This work addresses robot navigation under continuous observations and bounded continuous action spaces. Methodologically, we propose a message-passing-based autoregressive active inference agent: we formulate a factor graph model, decompose and distribute the expected free energy across a planning graph, integrate autoregressive modeling for online inference and action modulation, and dynamically balance exploration versus exploitation via predictive uncertainty. Compared to classical optimal controllers, our agent exhibits slightly slower convergence but achieves significantly improved accuracy and robustness of the learned dynamics model. To our knowledge, this is the first approach to realize message-passing-based planning and adaptive action regulation within continuous domains using active inference principles. The framework establishes a scalable architectural foundation for Bayesian control in embodied agents, enabling principled, uncertainty-aware decision-making under continuous state-action spaces.

Technology Category

• Reasoning under Uncertainty: Sequential Decision Making
• Intelligent Robots: State Estimation
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Agentic search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

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.

Problem

• Designing message passing inference for autoregressive active agents
• Validating agent performance in robot navigation tasks
• Comparing agent behavior with classical optimal controllers

Innovation

• Message passing on factor graph for inference
• Distributed expected free energy in planning
• Modulates actions based on predictive uncertainty