Black Friday Sale

Technology

Construction robot builds massive stone walls on its own

A robotic construction vehicle can use 3D digital mapping and AI to learn the best placement for each stone in a wall, without instruction from a human

By Jeremy Hsu

22 November 2023

A robotic construction vehicle picks and scans each boulder to determine where it fits in a stone wall

Marc Schneider

An autonomous robot with a large gripper can transform a pile of boulders into huge stone walls without mortar – learning on its own how to place each irregularly-shaped stone as the next building block.

The robotic excavator has built a stone wall 6 metres high and 65 metres long through a public park on the outskirts of Zurich, Switzerland. It also used a large shovel to autonomously landscape the park’s terrain into terraces.

“This is the first work to apply such a robotic excavator for the large-scale construction of permanent dry stone walls,” says Ryan Luke Johns at ETH Zürich in Switzerland.

Johns and his colleagues equipped the robot with lidar, which employs lasers to measure distances, so it could create its own 3D map of a construction site. They also trained several artificial intelligence models to help the robot figure out the best way to grasp and place individual stones.

Once the robot knows the location of each large stone based on the digital map, it grasps the rock, digitally scans it and determines its weight to get a better sense of how it may fit into the eventual wall.

Sign up to our The Weekly newsletter

Receive a weekly dose of discovery in your inbox.

After this scanning process, the robot can place one stone building block every 12 minutes. That is still 10 per cent slower than experienced human machine operators, who have an average rate of 11 minutes per stone placement. But manual construction requires additional workers to put down visual markers such as paint and string as guides for the operator.

The construction of walls without mortar also typically requires workers on foot to use shovels or their hands to add supporting stones, gravel and dirt. By comparison, the robot’s digital mapping and AI-powered analyses allow it to place stones almost perfectly most of the time, with a median positional error of just one-tenth of a metre – although its 82 per cent success rate for grasping stones sometimes led to follow-up grabbing attempts.

A human still supervised the robot, and a person drove it between locations on the active construction site for safety reasons when it was operating near human workers and other machines. But the researchers aim to figure out how the robot can operate fully autonomously and safely alongside others.

Journal reference:

Science Robotics DOI: 10.1126/scirobotics.abp9758

Topics: