Luis A. Mateos

Pipe joint rehabilitation - Short video showing only key elements of the process, cleaning and sealing the pipe-joint.

Fresh water pipe joint rehabilitation steps with a virtual in-pipe robot.


TITLE: Never send a human to do a machine's job:

Solving the fresh water leakage problem with semi-autonomous in-pipe robots

ABSTRACT: The limits of in-pipe robotics were explored by rehabilitating a segment of fresh-water pipe with a single monolithic robot.

Fresh water pipe leakage is considered to be a major problem, wasting a vital natural resource and damaging the supplying systems and foundations of roads and buildings (1). Governments around the world are tackling this problem with a variety of solutions, from humans inside the pipes, inner-layers of polyether (2) to robots .

In-pipe robotics have been so far restricted to be remote controlled without embedding AI as a core (3). A main reason is that complex manipulation with repairing tools inside a fragile pipe requires mastering skills, such as force control, visual perception and human consideration, plus managing their complex interactions.

We wanted to explore whether, based on the in-pipe robotics state-of-the-art, it is possible to tackle with one semi-autonomous robot a full pipe-joint rehabilitation that requires multiple specialized in-pipe robots or all human skills.

Unlike previous works (4), to emphasize the genericity of our robotic setup, we integrated in the robot only commercial hardware and common hand-held power tools: the robot is assembled with aluminum profiles and integrates waterproof gas springs, linear actuators, thin film force sensors and four cameras.

We achieved the semi-autonomous pipe-joint rehabilitation as shown in the accompanying video (movie S1). The test was performed in an one-meter diameter fresh water pipe segment in the city of Vienna, Austria with the following steps: (i) detection of the pipe-joint and positioning of the robot in the working area, (ii) extend the wheeled-legs to create a rigid structure inside the pipe so the spherical arm with the tool rotates centered in cylindrical 3D space, (iii) clean the pipe-joint with power grinding tools, (iv) seal the pipe-joint with an epoxy recipe. Overall, the pipe rehabilitation of a single highly corroded pipe-joint took 164 min 47 s, of which 23 min 13 s for robot insertion, pipe-joint detection and positioning, 87 min 22 s for fully cleaning the pipe joint with grinding tools, 54 min 7 for epoxy preparation and sealant application.


The robot moves in a differential wheel drive fashion to promptly adjust its horizontal position by an angle follower algorithm that minimizes the inclination degree when moving in the curved pipes surface. The challenge is to compensate the inconsistent estimated position of the robot over time to obtain a reliable odometry.

The pipe-joint detection was achieved by analyzing video frames from its omni-directional camera, applying a matching algorithm to find the pipe-joint pattern. The challenge is to reliably detect the pattern with a sensor fusion scheme of visual odometry (5), since the pipe-joint can be as small as 2 mm in highly corroded areas.

The positioning of the robot to the center of the pipe was performed by an automatic centering algorithm that simplifies the 3D space of the robot in a couple of 2D planes (6), each representing one set of wheeled-legs. The challenge is to consistently scan the surrounding pipe surface to find the optimal center position, even if the legs of the robot are stepping on corroded tubercles.

The tool positioning was performed by a couple of spherical robotic arms opposite to each other, located at the front of the robotic unit. One integrates the tool to clean or seal and the other integrates a driving wheel to move them in cylindrical 3D space covering the entire pipe-joint area (7). The challenge is to automatically control the driving wheel from its Dynamical Suspension Systems (DSS) to maintain a proper grip to the pipe.

The cleaning and sealing operations involve contact between the robot and the pipe surface that requires to be regulated. This is challenging because the vibrations and jump-back forces from the power tool affects its precise positioning. Even though the spherical robotic arm with the tool integrates a DSS, we equipped the robot with thin film pressure sensors mounted in the robot legs and implemented an indirect force control method (8). The challenge is to quickly and consistently measure the pressure on the legs to promptly adjust its position as a human does when handling power grinder.

The forces obtained from the sensors are converted, through a linearized proportional block to a position error, which is then fed to the robot’s position controller. In this sense, we precisely and consistently detect back forces from the power tool and adjust the centered position from the robot to comply with the tolerances for the smallest pipe-joint dimension.

Adaptability to fit in a wide range of pipe diameters is an extreme challenging task for in-pipe robotics because the robot requires to compress all rigid components (9). We extended the reconfigurable capabilities of rigid bodies in a constrained geometry by incorporating a latching mechanism based on motorized screw to fold aluminum profiles and then reconstruct them without losing the payload capacity inside the pipe (10).


We have shown that our developed in-pipe robot can achieve a highly complex task: rehabilitating a leaky pipe-joint in a fresh water pipe segment by leveraging existing developments in vision, control and automation.

The whole rehabilitation process was performed using commercial hardware and generic power tools, demonstrating its portability and capability to work in real environment. This capability could unleash the full potential of a new generation of autonomous in-pipe robots to address the needs for improving the pipe leaking problem world-wide.

There are still important limitations: Although all the steps were planned and controlled, their sequence was hard-code through a considerable engineering effort. Also, the epoxy application played a timing constrain in the process.

One can envision such a sequence being automatically determined by the autonomous robots with minimal interaction with a human supervisor or, ultimately, from a learning expert system. Combining the capabilities and the framework developed here with the recent advances in AI could lead, in the near future, to fully autonomous robots for pipe maintenance and rehabilitation.


Movie S1. Pipe joint rehabilitation - Video showing only key elements of the process, cleaning and sealing the pipe-joint.
Movie S2. Fresh water pipe joint rehabilitation steps with a virtual in-pipe robot.


1. S. Burn, D. Desilva, M. Eiswirth, O. Hunaidi, A. Speers, J. Thornton, Pipe Leakage - Future Challenges and Solutions, in Pipes Wagga Wagga Conference 1999, pp. 263 -270.

2. V. Archodoulaki, G. Kuschnig, S. Lueftl, M. Werderitsch, Silane modified polyether sealant failure in drinking water pipes, in Proceedings of the 2010 International Conference on Modification, Degradation and Stabilization of Polymers, 5 to 9 Sep 2010, Athens, Greece, pp 97 - 101.

3. H. Choi, S. Roh, in Bioinspiration and Robotics: Walking and Climbing Robots, (InTech 2007) pp. 375–402.

4. P. Dhananchezhiyan, S. S. Hiremath, M, Singaperumal, R. Ramakrishnan, Design and development of a reconfigurable type autonomous sewage cleaning mobile manipulator. in Proceedings of the 2013 International Conference on Design and Manufacturing, pp. 1464 – 1473.

5. P. Hansen, H. Alismail, P. Rander, B. Browning, Monocular visual odometry for robot localization in LNG pipes, in Proceedings of the 2011 IEEE International Conference on Robotics and Automation, 9-13 May 2011, Shanghai, China, pp 3111 - 3116.

6. L. A. Mateos, M. R. Dominguez, M. Vincze, Automatic in-pipe robot centering from 3D to 2D controller simplification, in Proceedings of the 2013 IEEE Intelligent Robots and Systems, 3 to 7 Nov 2013, Tokyo, Japan, pp. 258–265.

7. S. Kucuk, Z. Bingul, The inverse kinematics solutions of industrial robot manipulators, in Proceedings of the 2004 IEEE Mechatronics, pp. 274 – 279.

8. J. Park, D. Hyun, W. Cho, T. Kim, H. Yang, Normal-Force Control for an In-Pipe Robot According to the Inclination of Pipelines. IEEE Transactions on Industrial Electronics. 58, 12, 5304 – 5310 (2011).

9. T. li Yang, A. xin Liu, Ma, L.-Z, Hang, L.-B, Structure composition principle of reconfigurable mechanisms and basic methods for changing topological structure, in Proceedings of the 2009 ASME Reconfigurable Mechanisms and Robots, pp. 104 –109.

10. L. A. Mateos, M. Vincze, LaMMos - latching mechanism based on motorized-screw for reconfigurable robots, in Proceedings of the 2013 IEEE Advanced Robotics, 25 to 29 Nov 2013, Montevideo, Uruguay, pp. 1–8.

2019 Luis A. Mateos