Side-scan sonar mapping of an unknown large-scale seafloor area by a marine vehicle is nowadays very common. It is also important that a-priori unknown interesting parts of the seafloor area are scanned in more detail, i.e. sonified from both sides. However, completely autonomous and time-efficient coverage path (re)planning for such missions is still an open issue. In contrast to the standard overlap-all-sonar-ranges lawnmower pattern offline static coverage problem solution for side-scan sonar missions, in this paper two online sonar data-driven coverage algorithms are proposed as extensions of authors’ prior work. Analytical upper and lower bounds on performance of the proposed coverage planning algorithms are given and validated through extensive mission parameters variation simulations. Statistical performance analysis of the proposed coverage planning algorithms’ performance shows significant complete coverage time efficiency improvements w.r.t. the classical unadaptive lawnmower approach. Also, a detailed comparison of coverage planning algorithms proposed by the authors so far is provided.

A modular measurement model Extended Kalman filter (EKF) for for unmanned underwater vehicle (UUV) localization is proposed. Except for using measurements from UUV’s sensors, this EKF is augmented by ultra-short baseline range and visual-data based localization from an unmanned surface vehicle, and in-sonar image estimated UUV position. It is shown that the proposed EKF significantly enhances UUV’s navigational accuracy through a collaborative fusion of sensor data from multiple heterogeneous marine vehicles. Also, an Extended Rauch-Tung-Striebel (ERTS) smoother was run aposteriori to further improve UUV’s localization, which is shown to be very useful for accurate post-processing of the data acquired by the UUV.

Mapping an unknown large-scale marine area by a side-scan sonar onboard a marine vehicle as quickly as possible is often of great importance. It is also important that a-priori unknown interesting parts of the area are scanned in more detail, i.e. with the removal of sonic shadows. In contrast to the standard overlap-all-sonar-ranges lawnmower pattern, which is an offline static coverage problem solution for side-scan sonar missions, here a novel online side-scan sonar data-driven coverage solution is proposed. The proposed coverage algorithm provides a coverage solution based on local information gain from side-scan sonar data. At the same time, the solution is generated in such a way that coverage path length is minimized while covering the same area as the standard lawnmower. Upper and lower bounds of the proposed algorithm's improvement compared to the overlap-all-sonar-ranges lawnmower method are estimated analytically and validated through extensive mission parameters variation simulations. Simulation results show that our approach can cut down coverage path length significantly compared to the standard lawnmower method in most application cases.

A modular open source real-time simulation environment for linear model predictive control (LMPC) of line following at a constant depth is developed for underactuated marine vehicles. Two-stage tuning method of MPC's parameters is implemented as an extension of this simulator. MPC is used for high level control, i.e. setting the yaw rate reference, which is then tracked by the low level PID controllers. Simulation and experimental results show good performance of our MPC-PID control scheme when compared to ordinary PID controllers using Lyapunov-based guidance law.

Obtaining the optimal cost-to-go map for large scale rough terrains is computationally very expensive both in terms of duration and memory resources. A fast algorithm for approximation of the optimal cost-to-go map in terms of terrain traversability measures for path planning on known large scale rough terrains is developed. The results show that the majority of the cost-to-go map values, computed from every terrain location with respect to the goal location, are near-optimal. Unlike Dijkstra algorithm, the proposed algorithm has inherently parallel structure, and can be significantly speeded up depending on the number of used CPU cores.

It is well known that the process of tuning a fuzzy logic controller is almost always a very complex task, which is time consuming, very laborious and often requires expert knowledge of the controlled system. Mapping of fuzzy logic controller's parameters (rule base and membership functions of input(s) and output(s)) into a performance measure in a closed analytical form is near impossible to get, and thus the use of any classical optimization method is automatically ruled out. Knowing this, genetic algorithms with a fitness function in a form of cumulative response error represent a good choice of the optimization method. This approach enables the use of offline optimization of membership functions' parameters (which are being coded into chromosomes). Sugeno-Takagi fuzzy logic controllers with a proportional and a derivative component, and also with a fixed rule base are used in this approach. Experimental results of both simulations and validations on real systems are given in this paper and they show the good performance of this approach.