The existing huge need and large market for inspection and assessment of ageing transportation tunnels with visible signs of deterioration at a time that there are no funds or manpower to do the job, presents a big opportunity to the robotics industry. Towards this end, ROBO-SPECT proposes specialised work to adapt, extend and validate recent, exploitable, research results that will provide the required functionalities and intelligence to an automatic, multi-component and multi-degree-of-freedom robotic system that will credibly, efficiently and inexpensively perform both inspection and assessment of the tunnel in a single pass. Such a system will permit the wide spread deployment of robotic systems in the tunnel inspection market.

More specifically, what is needed is adaptation of recent research results to provide the required intelligence to identify defects of concern and the required intelligence to control the robotic arm to make measurements in the case of the latter defects. Needed are also adaptation/extension of existing sensors to provide measurements of defects of concern with the required accuracy and speed and algorithms that based on the latter measurements will assess the structural adequacy of the tunnel. To the above, one should add that state of the art navigation and positioning techniques should be adapted for tunnel inspection.

In this work the above will be provided in an automated robotic system for tunnel structural inspection and assessment that will be field evaluated in terms of its potential take-up and operational employment in a number of real cases.

The work will be carried out in 9 Work Packages (WPs) as follows:


All WPs receive input from WP9 Project Management.
All WPs provide input to WP8 Dissemination Activities and Exploitation of Results

WP1 (Scenarios and Metrics, User Requirements and System Architecture) will be led by EOAE. It involves identification of the needs of tunnel owners and operators, the inspection and assessment industry and the robotics industry and based on these, and taking into account the regulatory requirements, legal aspects and standards, the derivation of requirements and the design of the system architecture.

WP2 ( Robotic Navigation and Intelligent Positioning Controllers) will be lead by UC3M. A set of controllers will be developed for the robotic parts of the system (vehicle, crane, arm, global). Each controller will move a robotic part according to a reference it is given from a high-level (global) controller, but will provide basic security by restricting or suppressing the velocity of the movement if local sensor information indicates risk of collision. Part of the developments of WP3 will be used as the robotic perception system, as they will provide a stream of 3D models of the surrounding environment and the uncertainties of these models to the global controller. The main theory behind the global controller will be intelligent control, to enable it to accept the stream of models and uncertainties as perception input to update its belief state. The behaviours described in the first of the objectives in section 1.1 will be hard-coded into the global controller, and a prototype Human-Machine Interface will be developed for their individual or collective activation. The vehicle and crane will be developed by ROB, and the robot arm will be selected from the market.

WP3 (Computer Vision for Real Time Defect Detection) will be led by ENPC (along with ICCS). A set of local descriptors will be initially evaluated, which will be enhanced using semi-supervision tools. Object detection algorithms will be applied. Due to lack of data availability covering all the different and diverse types of tunnels, the focus lies on the use of semi-supervised learning methodologies for improving the recognition accuracy. In this case, approximate solutions are obtained using only the labeled data while accurate precision is drawn on the exploitation of a huge number of unlabelled data. For the recognition process, active continuous learning is performed so that cracks’ detection is on-line (real-time) accomplished as the system surveys the tunnels. The detection will be tailored to cylindrical coordination constraints to take care of the curved shape of the tunnel. Having recognized the objects, the next step of the vision tool will be to identify accurate 3D measurements. This will be achieved via monocular cameras yielding a cost tolerant methodology for the inspection. Smooth constraints on the shape of the detected objects will be imposed to handle the ill-posed problem of 3D measurements from 2D cameras. Finally, relevance feedback schemes will be adopted to enhance recognition accuracy exploiting user’s evaluation regarding the actual assessment. Relevance feedback is an online learning strategy which adapts the response of the system by exploiting user interaction. Furthermore relevance feedback schemes do not force the user to acquire expert technological knowledge about the structures, relations and decisions that the system applies in order to identify events of interest and alert. Instead, the schemes just receive an evaluation of the system performance, i.e., what is false positive or false negative, and then the schemes applies intelligent learning algorithms to modify the decision rules and conclude to a concrete, user-centric judgment.

WP4 (Measuring Devices for the Width and Depth of Cracks) will be led by CNR and it mainly involves the design and assembly of a sensor system prototype suited for integration on the robotic platform for measuring the width and depth of cracks on-field.

The objective in WP5 (Structural Assessment of Tunnels based on Inspection Results) is to adapt previous research on linings where structural assessment was based on monitoring of the both the extrados and intrados to the case where only the lining intrados is being inspected. It involves DBA, TECNIC and RISA (the WP leader).

The work in WP6 (System Integration and Validation in the Lab) involves integration of the systems in WPs 2, 3, 4, and 5. An integration platform will be set-up from the architecture works in WP1 and will integrate step by step all the components. A first phase will validate the interfaces between components and a second phase will perform the functional integration to qualify the system according to the performances specified in WP1. The integration platform will be supported by the existing CAS simulation platform that encompasses the environment simulation and the capability to play realistic scenarios. WP6 will also prepare/validate in factory the configuration for the field tests of WP7

The aim in WP7 (Testing, Field Evaluation and Benchmarking), led by EOAE, is testing, evaluation and benchmarking of the robotic system in WP6 at an early stage at the research infrastructure of tunnels at VSH and at a later stage at three actual road tunnels and a rail tunnel in terms of its potential take-up and operational employment.

WP8 (Dissemination Activities and Exploitation of Results) involves dissemination of the project’s results (VSH and RISA will be the leaders) and exploitation of the results (CAS will be the leader).

WP9 (Consortium Management) involves the overall project administrative and financial management at the level of the consortium and will be led by ICCS.

This project is funded by the European Union