Autonomous flight capabilities for inspection of cargo holds
This aerial drone, developed by the University of the Balearic Islands, is intended to address the operational scenario involving wide volumes with a reduced number of obstacles and irregular surfaces, like cargo holds and cargo tanks. The platform is characterized by a rich equipment of sensors and software technology for highly autonomous unsupervised navigation, able to explore wide spaces according to complex path planning.
The drone developed by the UIB team addresses the visual inspection of vessel cargo holds. To this end, it has been configured, in terms of equipment, locomotion abilities and operational features, to satisfy the scope of the survey activities, aiming at safer, more cost-efficient and more intensive visual inspection of wide spaces. Starting from the aerial platforms developed along previous inspection-related projects in which the UIB team has participated, and taking into account the lessons learnt during the respective field trials, several options about structure, control approach, platform localization, inspection-oriented capabilities to incorporate and interface with the user have been considered, solved and integrated to increase the Technology Readiness Level (TRL) of the whole inspection solution. For a start, a multi-rotor platform has been used as a basis of the mechanical structure, being fitted with a sensor suite intended to provide the pilot with enhanced functionality and autonomy during inspection flights, and as well supply the required inspection data, i.e. imagery, from the relevant areas.
In order to provide the platform with the necessary capabilities, a control architecture has been specifically designed and developed around the Supervised Autonomy (SA) control model, by means of extensive use of behaviour-based high-level control (including obstacle detection and collision prevention), all specifically devised for visual inspection. Accordingly, after analysing the visual inspection application, a number of suitable behaviours oriented towards visual inspection have been defined and integrated.
Figure 2 – High-level view of the inspection-oriented control architecture
The aerial platform is complemented with a Base Station (BS), comprising a Ground Control Unit (GCU) and suitable Hardware Interaction Devices (HID) to facilitate the interaction with the MAV and as well collect and process inspection data. Special emphasis has been put on the design of the BS to shorten the deployment times once in the field.
Usability and assistance to inspection operations, and how the supervised autonomy paradigm can contribute to both, have been key issues along the different developments that have been undertaken, either regarding the platform itself or the BS, and at all levels, hardware and software. Successful performance, in the form of both qualitative and quantitative results, has been reported for a set of experiments aiming at assessing the different functionalities of the platform. The full system has been evaluated both within the laboratory and out of the laboratory, on-board real vessels. The laboratory experiments have allowed us to check in detail and accurately the different functionalities of the platform, both at the scale of the laboratory and at a wider in-campus facility that has allowed us to continue the systematic assessment of the platform. On the other side, field trials, with the presence of surveyors guiding the experimentation, have permitted evaluating the platform under conditions closer to the final use. Results show the platform effective for visual inspection due to the inspection-oriented capabilities it has been fitted with.
Figure 3: Pictures from field trials on-board a bulk carrier (bottom: plot of 18 x 6 m sweeping performed in a cargo hold)
Expected progress beyond state-of-the-art
The focus will be on addressing those issues that can prevent the incorporation of these vehicles in routine cargo hold inspections and adopting specialized strategies for each issue that typically arises in this kind of environment.
Essentially, the following problems require specific answers that, once attained, will take us beyond the state of the art in vessel inspection:
Although being large areas in general, operating conditions in cargo holds can change significantly from one case to another, from poorly lit (or even completely dark) environments with a single entry point (man-hole sized) in oil tankers to open, wide areas under sun light in the case of bulk carriers (hatch door open) and containerships. This makes MAV navigation a challenge since a single sensor modality will not always be appropriate: e.g. dark environments do not permit using navigation solutions based on cameras, but other sensors, such as laser-, infrared-, sonar- or radar-based devices must be adopted; featureless scenarios also limit the effectiveness of vision-based motion estimation methods, while long corridors or walls (in comparison with sensor maximum range), where the environment does not present a discriminative shape that can be matched uniquely across scans, puts laser-based solutions in trouble, etc. Moreover, accurate motion estimation is not only necessary for platform control, but also to tag collected inspection data with precise positioning information. In order to counteract this problem (not solved yet for the case of vessels inspection), a two-fold hybrid solution will be adopted: on the one hand, a combination of relative and absolute localization will be considered, e.g. laser-based navigation together with easy-to-deploy UWB technology; on the other hand, the MAV is intended to be designed around a modular, flexible, and easily exchangeable sensor suite (both HW and SW levels), so that it can be adapted to the environment where the platform is required to operate, e.g. laser-based against vision-based navigation.
Collected inspection data must be complete and be properly presented in order for the user to do the most of it. In this regard, we plan to ensure proper coverage of the structure under inspection, e.g. a bulkhead, as well as to facilitate the integration of these data into a specific framework for enhanced visualization, e.g. 3D view rendering.
Being one of the main goals the incorporation of these platforms into routine inspection procedures, platform usability and robustness become key aspects. In order to do so, a supervised autonomy approach will be adopted. Under this paradigm, the robotic platform implements a number of behaviours that allows it to attain in a modular way different levels of autonomy and cover different kinds of tasks and situations, ranging from low- to medium-level complexity, without human intervention. The idea is not to switch from one mode of operation (total autonomy) to the other (teleoperation), but to implement a global approach under which a man is always in the loop, monitoring the progress of the task under execution, which if possible runs in autonomous mode, but allows him to take control whenever necessary. Besides, thanks to the behaviours running onboard the platforms, the operator does not need to deal with (nor to be trained about) the control complexities of the platforms, but can interact through terse, intuitive commands. In this way, the full system can attain a high level of robustness since simple situations are solved by the robotic platforms autonomously (i.e. the ones that the platforms can work out reliably), while those of high complexity are supported by an operator through a friendly, useful interface. Consequently, situation awareness, human judgment and decision making takes place at the precise level where they are required, and with the proper support through the available technological means.
Usability not only refers to the development of software that implements the functions actually needed by the user, but also means hardware that facilitates, in this case, the inspection procedures. To this end, as mentioned above, we plan to proceed case by case, and adopt a modular structure, which permits exchanging components depending on the particular inspection to perform. This is to make compatible the different requirements with the payload restrictions. As well, this will mean a challenge, from not only the mechanical/electrical point of view, but also regarding the Control Software Architecture (CSA). It will need to be able to adapt to a number of configurations, both regarding the interaction with the current hardware and regarding the capabilities which that hardware fits the platform with. Easy transport and deployment are additional desirable features that have an impact on usability which will also have to be dealt with: e.g. easy deployment of UWB infrastructure, practical transport of the platform itself, etc.
State of the art
A number of recent works have considered the use of Micro-Aerial Vehicles (MAVs) within the context of the inspection and monitoring of industrial facilities and assets, for data collection at remote or safety-compromised areas, difficult to reach by humans and ground vehicles, and with large areas to be covered as fast as possible. The aforementioned works consider, among others, power plant boilers, dam walls and penstocks, bridges, power lines, wind turbines, mines and tunnels, petrochemical facilities and large-tonnage vessels.
In order for these platforms to be of practical use in the aforementioned cases, typically GPS-denied scenarios, sufficient onboard sensors and computing power are needed to stabilize the platform and localize it within the environment, build maps, perceive and avoid obstacles, and plan flight trajectories. Solutions published so far mainly differ in the sensor(s) used, the amount of processing that is performed onboard/off-board and the assumptions made about the environment. Among the different sensing devices available, the following reviews those that can be of use for the case of cargo hold inspection given its character of rather larger, regular-shaped areas: 2D laser scanners, Vision-based navigation, Depth and RGB-D cameras.
For the particular problem of pose estimation (i.e. not including collision detection), wireless-based localization has recently received a significant amount of attention as an alternative of GPS in poor signal reception areas, being a priori especially suited for the kind of scenarios which can be found in cargo hold vessels. Among the different possibilities available nowadays, Ultra-Wide Band (UWB) systems have emerged as one of the leading positioning technologies because the UWB ultra-short pulses are resilient to frequency-dependent absorption, thanks to their large bandwidth, and because ultimate accuracy can range from 2 cm (ideal conditions) to 50 cm (non-line of sight scenario). Several categories of UWB algorithms can be distinguished depending on how the position is inferred from the radio signals travelling between the beacons and the target node: based on Time of Arrival (TOA), based on Time Difference of Arrival (TDOA), based on Angle of Arrival (AOA), based on Received Signal Strength (RSS), and hybrid approaches. TOA and TDOA have been shown to exhibit higher accuracy than the other options, and even better results have been achieved by the hybrid approaches that combine some of the different alternatives.