Active Inference control is a novel control method based on the free energy principle, which combines action, perception and learning [1][2]. The first Active Inference controller showed promising results on a 7-DOF robot arm for a pick and placing task, however it took nearly six seconds to converge which is too slow [2]. This thesis aims to reduce the convergence time of an Active Inference controller. Therefore an Active Inference controller that is used to control the velocity of a Jackal robot was developed. It was compared to the standard differential drive Controller and it was found that the standard controller outperformed the Active Inference controller on both rise and settling time. A state space model was derived to obtain a better understanding of the Active Inference controller, for this model the robot was modelled as a point mass. This assumption was confirmed by a step response test of the robot, placing it both on the box and on the ground. It was found that both behaved similar, and thus the wheel ground interactions and internal dynamics of the robot did not affect the convergence time. The state space model was used to find three methods to reduce the convergence time of the Active Inference controller. As a first method the update frequency of the entire controller was increased, for the second method only the update frequency of the inner belief update loop was increased. The last method used an increased update frequency of the belief update loop and the parameters of the system were tuned. The methods were tested using step response tests and a square wave test, first in a simulated Gazebo environment and later with the actual robot. It was found that the third method reduced the converging time of the Active Inference controller, and reduced it from 0.88s to 0.51s. The improved controller was also tested against the differential drive controller outperformed it. It was found that due to the increased update frequency the tuning parameters could be changed to a wider range of values, this resulted in a shorter convergence time.