The optimal power flow (OPF) problem is a classic and widely-studied topic in the field of power systems.Its purpose is to minimize the running costs of a power system by determining the optimal operating points,while respecting a set of physical constraints. While most power systems are currently controlled by a centralcoordinator, there is significant interest in the research of decentralized schemes. By eliminating the need fora central coordinator, fully decentralized algorithms eliminate the single point of failure, thereby enhancingthe reliability of the overall system. Moreover, as communication and telemetry play a progressively vital rolein the power grid of the future, decentralized methods reduce the strain on the data infrastructure. In otherwords, the computational and communication load on the central computer is decreased, as it no longer hasto gather and process large amounts of information from all the nodes in the system. As a result, although newchallenges are introduced with distributed algorithms, the OPF becomes more scalable in some regards. Onefully decentralized method for performing OPF calculations is Consensus+Innovation (C+I). The C+I methodis based on the Karush-Kuhn-Tucker (KKT) Conditions class of optimization, where the solution is founditeratively using an update strategy. Moreover, communication is only required between directly connectednodes, or neighbors. This method has been applied in the cases of AC systems as well as unipolar DC systems,reporting promising results. However, until now, no decentralized algorithm existed for bipolar DC systems.Bipolar topologies, while they are more complex to model and optimize, possess numerous advantages overunipolar ones. By providing two voltage levels, larger loads can be connected directly between the two polesin order to provide double the power. This increased voltage level allows for the current to be halved for thesame amount of power, thereby significantly reducing the conductive losses. On the other hand, smaller loadsare connected between one of the poles and the neutral, meaning that power electronic devices with lowerratings can be used. Furthermore, generators and loads can be connected in parallel at the same location.This means that they must be modelled as controlled current sources which are connected between twophysical nodes. In other words, the optimization variables are altered from nodal voltages and powers tonode voltages and source currents. However, these changes in the modelling, optimization, and control of thegrid mean that the single line diagram (SLD) equivalent circuit, used by the algorithms for unipolar systems,is no longer valid for bipolar ones. Thus, the objective of this thesis was to develop a fully decentralized OPFalgorithm for bipolar DC grids. The algorithm is based on online optimization, meaning that measurementsfrom the physical grid are taken and used in every iteration of the OPF. Moreover, the algorithm is based onbipolar DC grids wherein the power electronic converters follow a droop control scheme, the effect of whichis directly accounted for in the newly developed update strategy. The algorithm is tested on four differentsimulated case studies with varying operational scenarios. These test cases include both fixed and variable-power loads and controllable generators. Furthermore, the algorithm is demonstrated to successfully achieveline congestion management using demand response. Finally, a case with unbalanced loading is simulatedsuccessfully, while respecting all of the physical limits of the system.