Active inference for Robot control: A Factor Graph Approach


  • Mees Vanderbroeck Delft University of Technology
  • Mohamed Baioumy Delft University of Technology
  • Daan van der Lans Delft University of Technology
  • Rens de Rooij Delft University of Technology
  • Tiis van der Werf Delft University of Technology



Factor graphs, free-energy principle, active inference, closed-loop control, variational message passing


Active Inference provides a framework for perception,
action and learning, where the optimization is done by
minimizing the Free-Energy of a system. This paper
explores whether active inference can be used for closedloop
control of a 1 degree of freedom robot arm. This is
done by implementing variational message passing on
Forney-style factor graphs; a probabilistic programming
framework. We show that an active inference controller
with variational message passing can perform state
estimation and control at the same time.

Additional Files