Four-wheel inspection robot
- Self-centers in any orientation, no feedback needed
- Rolls a full turn to choose its branch at a T-joint
- Proven in steel pipe: straight runs, elbows and T-joints
Pipe Network Inspection Robot
Robots that inspect pipes from the inside.
Compact, self-centering robots for 2 to 6 inch pipe networks. They hold themselves against the wall, drive through elbows and T-joints, and are learning to find their way on their own.
Why it matters
Aging gas pipelines leak. Inspecting them means digging.
Most gas distribution pipe is buried under streets and buildings. A robot that can travel the network from the inside turns an excavation into an inspection.
1.18million km
Of oil and gas pipeline in operation worldwide, across more than 2,300 pipelines
Global Energy Monitor, 2020
2.35million miles
Of natural gas distribution mains and service lines in the United States alone
PHMSA annual reports, 2023
16,000miles
Of cast and wrought iron gas mains still in service in the US, some laid over a century ago
PHMSA, 2025
$7.75billion
Projected North American pipeline leak-detection market by 2035, from $3.76 billion in 2025
Oil and Gas Pipeline Leak Detection Market report, 2026
Gas lost through leaking distribution mains in the US was valued at about $192 million in a single year by the EPA's Inspector General, and in-pipe inspection has been estimated to avoid £14,000 per kilometer per year in replacement cost. Most of that pipe is 2 to 6 inches across, full of bends, valves and branches, and cannot be inspected by the tools built for large transmission lines.
EPA Office of Inspector General, 2014; National Grid, 2017.
State of the art
- Aging pipelines need inspection, maintenance and repair.
- Leaks are common, costly and dangerous.
- Conventional inspection robots are complicated and cannot steer through junctions.
Our robots
- Stable, simple and robust, built from off-the-shelf parts.
- Maneuverable across pipe diameters, orientations and joint types.
- Learning to navigate pipe networks on their own.
How it works
Let the spring do the holding.
Four wheels in one plane. Spring-loaded arms press the wheels against opposite walls, so the robot centers itself in the pipe without any feedback.
-
Drive
Two omniwheels at the ends of the arms push the robot along the pipe, forward or back.
-
Hold
At each shoulder a servo winds a torsion spring. The spring, not the motor, presses the arm into the wall, so the grip stays steady through bumps, joints and changes in diameter.
-
Roll
Two control wheels set across the pipe axis rotate the whole robot, so it can line up with the branch it wants at a T-joint before it turns.
Driving the length of a 6-inch steel test pipe, supplied by GTI Energy, and back.
The robots
One idea, two robots.
The same self-centering design scales from 4 to 6 inch mains. A smaller robot covers the 2-inch lines that run from mains into buildings.
Miniature robot
- Two arms, three omniwheels, one spring
- Climbs vertical pipe and takes elbows and U-bends
- Controlled from a phone over Bluetooth
Four-wheel robot
In the steel pipe.
The 6-inch robot has been run through straight sections, elbows and both ways through T-joints in a steel test network with real oxidation and weld gaps: along the primary conduit and into the lateral branch.
The 4-inch version shares the same layout and actuator.
Miniature robot
Two inches of pipe, three wheels.
One motor drives two omniwheels through a brass gear train; a third wheel on the second arm steers. A torsion spring at the joint between the arms presses all three wheels into the wall, which is enough grip to climb vertical pipe.
Tested in horizontal and vertical runs, elbows, a 9.5-inch bend and U-shaped pipe.
Autonomy
Reading the pipe from the inside.
Underground there is no GPS and no map. A single fisheye camera tells the robot what joint it is looking at, how the joint is oriented, and how far it has traveled.
99%+
Agreement between the vision-estimated joint angle and the IMU reference
5
Joint types recognized: straight, connector, elbow, T primary, T branch
0.9%
Position error over a 30 cm run with camera and IMU fused
1
Camera. Everything runs on a Raspberry Pi 5 on board, with no GPU
The method fits arcs to the shadow and edge of the joint and is built to tolerate rust, debris and dented walls, because that is what real steel pipe looks like. Distance comes from visual-inertial odometry: a laser ring gives the camera a known scale from the pipe's own diameter, optical flow tracks the oxidation texture on the wall, and an IMU carries the estimate through stretches that are too dark or too blurred to trust.
Timeline
From a concept to a robot in real steel.
Phase 12022 to September 2026
- 2022 to 2023
A stable design for unstable places
Funded by ERDC-CERL through GTI Energy to inspect buried gas pipe networks. The lab chooses a stable, spring-loaded, planar design and files a patent application through UC San Diego.
- 2024
Actuator and first robots
The torsion-spring series elastic actuator is presented at ICCR in Lyon and wins the best oral presentation award. The 4-inch and 2-inch robots are built and tested.
- 2025 to 2026
Steel pipe and autonomy
A 6-inch version runs in a steel test network with real joints. Joint classification, orientation estimation and visual-inertial odometry run on board.
Phase 2From October 2026
- Now
Closing the loop
Camera on the robot, autonomous traversal of a network, and planning for deployment with the project's sponsors.
Publications
Papers and code.
- 2026Vision-Based and Multi-Task Deep Learning for Steel Pipe Classification and Orientation Estimation in Autonomous Pipeline Inspection Systems
- 2026Series elastic actuation enables passive self-alignment in multi-planar pipe networks
- 2025
- 2024A new robot design for the interior exploration and inspection of pipe networks
- 2024Design and Characterization of the Torsion Spring-Motor Integrated Series Elastic Actuator
Patent pending: UC San Diego case SD2023-185, a new class of stable, spring-loaded, planar, roll-controlled robots for the interior exploration of networks of pipes and ducts.
Team
Who built it.
- Thomas BewleyPrincipal investigator. Professor, Mechanical and Aerospace Engineering
- Tao-Yi WanProject lead. Postdoctoral researcher; mechanisms, vision and navigation
- Joonyoung JangFour-wheel robot design and series elastic actuator. PhD 2024
- Viswa SubramanianLearning-based joint classification
Sponsors
Supported by
This research is funded by the ERDC Construction Engineering Research Laboratory (CERL) of the US Army Corps of Engineers and by GTI Energy, to whom the lab is grateful for their generous support, for the steel pipe test network, and for the problem itself.
Contact
Thomas Bewley, tbewley@ucsd.edu
Tao-Yi Wan, twan@ucsd.edu
Coordinated Robotics Lab, Department of Mechanical and Aerospace Engineering, UC San Diego, 9500 Gilman Drive, La Jolla, CA 92093.