Message passing-based inference in an autoregressive active inference agent
This is an incremental improvement for robotics, focusing on exploration and exploitation in continuous-valued spaces.
The paper tackled the problem of designing an active inference agent for robot navigation by using message passing on a factor graph, resulting in the agent arriving later but with a better model of dynamics compared to a classical optimal controller.
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.