Moving towards a world with ubiquitous automation, efficient design of intelligent autonomous agents that are capable of adapting to dynamic environments gains traction. In this setting, active inference emerges as a contender that inherently brings together action and perception through the minimization of a single cost function, namely free energy. Being a Bayesian inference method, active inference not only encodes uncertainty for perception but also for action and hence, it provides a strong expediency for building real-world intelligent autonomous agents that have to deal with uncertainties naturally found in the world. Coupled with other methods such as automated generation of inference algorithms and Forney-style factor graphs, active inference offers fast design cycles, adaptability and modularity. Furthermore, in cases where a priori specification of goal priors is prohibitively difficult, Bayesian target modelling opens up active inference to more complex problems and provides a higher-level means of speeding up design cycles through learning desired future observations. In order to assess active inference’s capabilities and feasibility for a real-world application, we provide a proof of concept that runs on a ground-based robot in order to navigate in a specified area to find the location chosen by the user through observing user feedback.