A Robotic Experimental Setup with a Stewart Platform to Emulate Underwater Vehicle-Manipulator Systems.

This study presents an experimental robotic setup with a Stewart platform and a robot manipulator to emulate an underwater vehicle-manipulator system (UVMS). This hardware-based emulator setup consists of a KUKA IIWA14 robotic manipulator mounted on a parallel manipulator, known as Stewart Platform, and a force/torque sensor attached to the end-effector of the robotic arm interacting with a pipe. In this setup, we use realistic underwater vehicle movements either communicated to a system in real-time through 4G routers or recorded in advance in a water tank environment. In addition, we simulate both the water current impact on vehicle movement and dynamic coupling effects between the vehicle and manipulator in a Gazebo-based software simulator and transfer these to the physical robotic experimental setup. Such a complete setup is useful to study the control techniques to be applied on the underwater robotic systems in a dry lab environment and allows us to carry out fast and numerous experiments, circumventing the difficulties with performing similar experiments and data collection with actual underwater vehicles in water tanks. Exemplary controller development studies are carried out for contact management of the UVMS using the experimental setup.

Robust Silent Localization of Underwater Acoustic Sensor Network Using Mobile Anchor(s).

Underwater acoustic sensor networks (UWASNs) can revolutionize the subsea domain by enabling low-cost monitoring of subsea assets and the marine environment. Accurate localization of the UWASNs is essential for these applications. In general, range-based localization techniques are preferred for their high accuracy in estimated locations. However, they can be severely affected by variable sound speed, multipath spreading, and other effects of the acoustic channel. In addition, an inefficient localization scheme can consume a significant amount of energy, reducing the effective life of the battery-powered sensor nodes. In this paper, we propose robust, efficient, and practically implementable localization schemes for static UWASNs. The proposed schemes are based on the Time-Difference-of-Arrival (TDoA) measurements and the nodes are localized passively, i.e., by just listening to beacon signals from multiple anchors, thus saving both the channel bandwidth and energy. The robustness in location estimates is achieved by considering an appropriate statistical noise model based on a plausible acoustic channel model and certain practical assumptions. To overcome the practical challenges of deploying and maintaining multiple permanent anchors for TDoA measurements, we propose practical schemes of using a single or multiple surface vehicles as virtual anchors. The robustness of localization is evaluated by simulations under realistic settings. By combining a mobile anchor(s) scheme with a robust estimator, this paper presents a complete package of efficient, robust, and practically usable localization schemes for low-cost UWASNs.

Online 3-Dimensional Path Planning with Kinematic Constraints in Unknown Environments Using Hybrid A* with Tree Pruning.

In this paper we present an extension to the hybrid A* (HA*) path planner. This extension allows autonomous underwater vehicle (AUVs) to plan paths in 3-dimensional (3D) environments. The proposed approach enables the robot to operate in a safe manner by accounting for the vehicle's motion constraints, thus avoiding collisions and ensuring that the calculated paths are feasible. Secondly, we propose an improvement for operations in unexplored or partially known environments by endowing the planner with a tree pruning procedure, which maintains a valid and feasible search-tree during operation. When the robot senses new obstacles in the environment that invalidate its current path, the planner prunes the tree of branches which collides with the environment. The path planning algorithm is then initialised with the pruned tree, enabling it to find a solution in a lower time than replanning from scratch. We present results obtained through simulation which show that HA* performs better in known underwater environments than compared algorithms in regards to planning time, path length and success rate. For unknown environments, we show that the tree pruning procedure reduces the total planning time needed in a variety of environments compared to running the full planning algorithm during replanning.

Learning Mobile Manipulation through Deep Reinforcement Learning.

Mobile manipulation has a broad range of applications in robotics. However, it is usually more challenging than fixed-base manipulation due to the complex coordination of a mobile base and a manipulator. Although recent works have demonstrated that deep reinforcement learning is a powerful technique for fixed-base manipulation tasks, most of them are not applicable to mobile manipulation. This paper investigates how to leverage deep reinforcement learning to tackle whole-body mobile manipulation tasks in unstructured environments using only on-board sensors. A novel mobile manipulation system which integrates the state-of-the-art deep reinforcement learning algorithms with visual perception is proposed. It has an efficient framework decoupling visual perception from the deep reinforcement learning control, which enables its generalization from simulation training to real-world testing. Extensive simulation and experiment results show that the proposed mobile manipulation system is able to grasp different types of objects autonomously in various simulation and real-world scenarios, verifying the effectiveness of the proposed mobile manipulation system.

3D Laser Scanner for Underwater Manipulation.

Nowadays, research in autonomous underwater manipulation has demonstrated simple applications like picking an object from the sea floor, turning a valve or plugging and unplugging a connector. These are fairly simple tasks compared with those already demonstrated by the mobile robotics community, which include, among others, safe arm motion within areas populated with a priori unknown obstacles or the recognition and location of objects based on their 3D model to grasp them. Kinect-like 3D sensors have contributed significantly to the advance of mobile manipulation providing 3D sensing capabilities in real-time at low cost. Unfortunately, the underwater robotics community is lacking a 3D sensor with similar capabilities to provide rich 3D information of the work space. In this paper, we present a new underwater 3D laser scanner and demonstrate its capabilities for underwater manipulation. In order to use this sensor in conjunction with manipulators, a calibration method to find the relative position between the manipulator and the 3D laser scanner is presented. Then, two different advanced underwater manipulation tasks beyond the state of the art are demonstrated using two different manipulation systems. First, an eight Degrees of Freedom (DoF) fixed-base manipulator system is used to demonstrate arm motion within a work space populated with a priori unknown fixed obstacles. Next, an eight DoF free floating Underwater Vehicle-Manipulator System (UVMS) is used to autonomously grasp an object from the bottom of a water tank.

Increasing circular synthetic aperture sonar resolution via adapted wave atoms deconvolution.

Circular Synthetic Aperture Sonar (CSAS) processing computes coherently Synthetic Aperture Sonar (SAS) data acquired along a circular trajectory. This approach has a number of advantages, in particular it maximises the aperture length of a SAS system, producing very high resolution sonar images. CSAS image reconstruction using back-projection algorithms, however, introduces a dissymmetry in the impulse response, as the imaged point moves away from the centre of the acquisition circle. This paper proposes a sampling scheme for the CSAS image reconstruction which allows every point, within the full field of view of the system, to be considered as the centre of a virtual CSAS acquisition scheme. As a direct consequence of using the proposed resampling scheme, the point spread function (PSF) is uniform for the full CSAS image. Closed form solutions for the CSAS PSF are derived analytically, both in the image and the Fourier domain. The thorough knowledge of the PSF leads naturally to the proposed adapted atom waves basis for CSAS image decomposition. The atom wave deconvolution is successfully applied to simulated data, increasing the image resolution by reducing the PSF energy leakage.

Underwater depth imaging using time-correlated single-photon counting.

A depth imaging system, based on the time-of-flight approach and the time-correlated single-photon counting (TCSPC) technique, was investigated for use in highly scattering underwater environments. The system comprised a pulsed supercontinuum laser source, a monostatic scanning transceiver, with a silicon single-photon avalanche diode (SPAD) used for detection of the returned optical signal. Depth images were acquired in the laboratory at stand-off distances of up to 8 attenuation lengths, using per-pixel acquisition times in the range 0.5 to 100 ms, at average optical powers in the range 0.8 nW to 950 µW. In parallel, a LiDAR model was developed and validated using experimental data. The model can be used to estimate the performance of the system under a variety of scattering conditions and system parameters.

Application of Adaptive and Switching Control for Contact Maintenance of a Robotic Vehicle-Manipulator System for Underwater Asset Inspection.

The aim of this study is to design an adaptive controller for the hard contact interaction problem of underwater vehicle-manipulator systems (UVMS) to realize asset inspection through physical interaction. The proposed approach consists of a force and position controller in the operational space of the end effector of the robot manipulator mounted on an underwater vehicle. The force tracking algorithm keeps the end effector perpendicular to the unknown surface of the asset and the position tracking algorithm makes it follow a desired trajectory on the surface. The challenging problem in such a system is to maintain the end effector of the manipulator in continuous and stable contact with the unknown surface in the presence of disturbances and reaction forces that constantly move the floating robot base in an unexpected manner. The main contribution of the proposed controller is the development of the adaptive force tracking control algorithm based on switching actions between contact and noncontact states. When the end effector loses contact with the surface, a velocity feed-forward augmented impedance controller is activated to rapidly regain contact interaction by generating a desired position profile whose speed is adjusted depending on the time and the point where the contact was lost. Once the contact interaction is reestablished, a dynamic adaptive damping-based admittance controller is operated for fast adaptation and continuous stable force tracking. To validate the proposed controller, we conducted experiments with a land robotic setup composed of a 6 degrees of freedom (DOF) Stewart Platform imitating an underwater vehicle and a 7 DOF KUKA IIWA robotic arm imitating the underwater robot manipulator attached to the vehicle. The proposed scheme significantly increases the contact time under realistic disturbances, in comparison to our former controllers without an adaptive control scheme. We have demonstrated the superior performance of the current controller with experiments and quantified measures.

Multiresolution 3-D reconstruction from side-scan sonar images.

In this paper, a new method for the estimation of seabed elevation maps from side-scan sonar images is presented. The side-scan image formation process is represented by a Lambertian diffuse model, which is then inverted by a multiresolution optimization procedure inspired by expectation-maximization to account for the characteristics of the imaged seafloor region. On convergence of the model, approximations for seabed reflectivity, side-scan beam pattern, and seabed altitude are obtained. The performance of the system is evaluated against a real structure of known dimensions. Reconstruction results for images acquired by different sonar sensors are presented. Applications to augmented reality for the simulation of targets in sonar imagery are also discussed.

The fusion of large scale classified side-scan sonar image mosaics.

This paper presents a unified framework for the creation of classified maps of the seafloor from sonar imagery. Significant challenges in photometric correction, classification, navigation and registration, and image fusion are addressed. The techniques described are directly applicable to a range of remote sensing problems. Recent advances in side-scan data correction are incorporated to compensate for the sonar beam pattern and motion of the acquisition platform. The corrected images are segmented using pixel-based textural features and standard classifiers. In parallel, the navigation of the sonar device is processed using Kalman filtering techniques. A simultaneous localization and mapping framework is adopted to improve the navigation accuracy and produce georeferenced mosaics of the segmented side-scan data. These are fused within a Markovian framework and two fusion models are presented. The first uses a voting scheme regularized by an isotropic Markov random field and is applicable when the reliability of each information source is unknown. The Markov model is also used to inpaint regions where no final classification decision can be reached using pixel level fusion. The second model formally introduces the reliability of each information source into a probabilistic model. Evaluation of the two models using both synthetic images and real data from a large scale survey shows significant quantitative and qualitative improvement using the fusion approach.

Human body pose estimation with particle swarm optimisation.

In this paper we address the problem of human body pose estimation from still images. A multi-view set of images of a person sitting at a table is acquired and the pose estimated. Reliable and efficient pose estimation from still images represents an important part of more complex algorithms, such as tracking human body pose in a video sequence, where it can be used to automatically initialise the tracker on the first frame. The quality of the initialisation influences the performance of the tracker in the subsequent frames. We formulate the body pose estimation as an analysis-by-synthesis optimisation algorithm, where a generic 3D human body model is used to illustrate the pose and the silhouettes extracted from the images are used as constraints. A simple test with gradient descent optimisation run from randomly selected initial positions in the search space shows that a more powerful optimisation method is required. We investigate the suitability of the Particle Swarm Optimisation (PSO) for solving this problem and compare its performance with an equivalent algorithm using Simulated Annealing (SA). Our tests show that the PSO outperforms the SA in terms of accuracy and consistency of the results, as well as speed of convergence.

Model-based approach to the detection and classification of mines in sidescan sonar.

This paper presents a model-based approach to mine detection and classification by use of sidescan sonar. Advances in autonomous underwater vehicle technology have increased the interest in automatic target recognition systems in an effort to automate a process that is currently carried out by a human operator. Current automated systems generally require training and thus produce poor results when the test data set is different from the training set. This has led to research into unsupervised systems, which are able to cope with the large variability in conditions and terrains seen in sidescan imagery. The system presented in this paper first detects possible minelike objects using a Markov random field model, which operates well on noisy images, such as sidescan, and allows a priori information to be included through the use of priors. The highlight and shadow regions of the object are then extracted with a cooperating statistical snake, which assumes these regions are statistically separate from the background. Finally, a classification decision is made using Dempster-Shafer theory, where the extracted features are compared with synthetic realizations generated with a sidescan sonar simulator model. Results for the entire process are shown on real sidescan sonar data. Similarities between the sidescan sonar and synthetic aperture radar (SAR) imaging processes ensure that the approach outlined here could be made applied to SAR image analysis.