Automated Planning, Sensing and Control for Autonomous Underwater Robotic Systems

Abstract

Today’s autonomous robotic systems have a significant impact on industrial applications and in academic disciplines. This dissertation considers a broad range of topics, from formal methods, sensor fusion, image processing, nonlinear control to controller synthesis of motion tasks with applications in underwater, mobile and underground robots. One of the most significant challenges in the robotics area lies in the area of motion and task planning. Motion planning is the robot able to move in the workspace while at the same time avoiding obstacles. On the other hand, task planning refers to the robot’s ability to execute a specific task in the workspace. The main aim is to be able for a given task in a high-level language the robot to compile this specification into low-level descriptions in order to accomplish a task. Autonomous underwater robots typically have to accomplish missions in an unknown and usually unstructured environment. The mission complexity grows considering to limited robot sensing systems as well as the limited on-line communications. For instance, GPS is not applicable due to the inefficient underwater electromagnetic transmission. In addition, vision-based systems are limited due to poor visibility in murky waters. The actuating system is usually composed of thrusters and control surfaces; all of them have non-linear dynamics and are strongly affected by the hydrodynamic effects. The ocean currents and flows imply additional difficulties for the ROV control system making the robot to deviate away from its desired state or path. Estimates of the flow velocity provided by various sensors or techniques may be incorporated into the control loop to compensate for the drift phenomenon. In this thesis, we address the problem of underwater visual inspection task as a combination of (i) a problem of localization and state estimation of the ROV with respect to the target by fusing information from different sources; (ii) a problem of control of an under-actuated underwater vehicle in the proximity to the target; (iii) a problem of full coverage of fishnet cages. Autonomous multi-agent coverage of large-scale under-ground sewer networks is also addressing in this dissertation. Sewer network systems are typically dendritic networks converging in the downstream direction without closed loops. In network systems theory such networks are characterized as a tree or more precisely directed tree networks where the directionality is inherited from the sewage flow direction. Sewer network flow channels dimensions are typically restricted, allowing only a single inspection robot at a given position. Robots operating in such networks can only interchange positions at channel junctions. Wireless communications in underground sewer networks are much more challenging than in above-ground settings. The main transmission path is through the underground network’s channels, usually non-line-of-sight and with severe attenuation over corridor bends and turns and there are also issues related to multi-path reflections. The performance of the proposed methodologies is verified in realistic simulations in 2D and 3D virtual environments. Furthermore, extensive experimental validation on the actual hardware was carried out in a controlled environment at Robotics, Control, and Decision Systems (RCDS) laboratory and in the field (open sea) under real conditions.