This report proposes three algorithms using model predictive control (MPC) in order to improve the positioning accuracy of an unmanned vehicle. The developed algorithms succeed in reducing the uncertainty in position by allowing the vehicle to deviate from a planned path, and can also handle the presence of occluding objects. To achieve this improvement, a compromise is made between following a predefined trajectory and maintaining good positioning accuracy.

Due to the recent development of threats to systems using global navigation satellite systems to localise themselves, there is an increased need for methods of localisation that can function without relying on receiving signals from distant satellites. One example of such a system is a vehicle using a range-bearing sensor in combination with a map to localise itself. However, a system relying only on these measurements to estimate its position during a mission may get lost or gain an unacceptable level of uncertainty in its position estimates. Therefore, this thesis proposes a selection of algorithms that have been developed with the purpose of improving the positioning accuracy of such an autonomous vehicle without changing the available measurement equipment. These algorithms are:

• A nonlinear MPC solving an optimisation problem.
• A linear MPC using a linear approximation of the positioning uncertainty to reduce the computational complexity.
• A nonlinear MPC using a linear approximation (henceforth called the approximate MPC) of an underlying component of the positioning uncertainty in order to reduce computational complexity while still having good performance.

The algorithms were evaluated in two different types of simulated scenarios in MATLAB. In these simulations, the nonlinear, linear and approximate MPC algorithms reduced the root mean squared positioning error by 20-25 %, 14-18 %, and 23-27 % respectively, compared to a reference path. It was found that the approximate MPC seems to have the best performance of the three algorithms in the examined scenarios, while the linear MPC may be used in the event that this is too computationally costly. The nonlinear MPC solving the full problem is a reasonable choice only in the case when computing power is not limited, or when the approximation used in the approximate MPC is too inaccurate for the application.