A four-legged robot that inspects concrete can walk through underground sewage tunnels and detect when they need repairing.
Hendrik Kolvenbach at the Swiss Federal Institute of Technology (ETH Zurich) in Switzerland and his colleagues have developed a that scratches one of its legs against concrete to determine the condition it is in. The robot is waterproof and it can wade through water and climb over obstacles.
Because many modern sewage systems were built decades ago, constant is needed to prevent major leaks. Currently, human inspectors do the job manually.
Advertisement
鈥淚t鈥檚 damp, it鈥檚 slippery, it鈥檚 dangerous, you have all sorts of rats,鈥� says Kolvenbach.
[video_player id=鈥漜94mf7te鈥� access_level=鈥漞veryone鈥漖
The robot, called ANYmal, is equipped with lidar and a stereo camera to help it position itself. Each of its limbs has three joints, enabling manoeuvrability.
Sensors built into its feet pick up information such as torque. When the robot scratches one of its feet against the concrete, it measures the vibrations generated, giving an indication of the roughness of the surface: the smoother the concrete, the better condition it is in.
The team tested ANYmal in sewage tunnels in Zurich, and first tasked it with inspecting an area where the concrete conditions had already been checked by human inspectors.
The robot took 625 samples of concrete conditions in areas where the concrete had been classified by humans as either good, satisfactory or fair. ANYmal didn鈥檛 collect information from areas where there were major deteriorations or leakages.
The team used this data to train a for the robot, which was able to distinguish between the three conditions with 92.6 per cent accuracy.
ANYmal was then deployed into an area that hadn鈥檛 yet been checked. It inspected the concrete floor every 2 to 3 metres, saving the information as a map. In total, the robot covered a distance of 300鈥塵etres, taking 130 inspection samples.
In future, the team plans to test ANYmal on other tasks in different environments, such as underground mines.
Journal of Field Robotics
