A robot that scans a construction site on its own
Published research2025ROS · LIO-SAM · Ouster OS1 · Leica BLK360 · SLAM
Problem
Recording what a construction site actually looks like today still means a person walking a scanner around it. That is slow and costly enough that it happens rarely, so the site record is usually out of date, and the decisions made against a stale record show up later as rework, delays and overruns.
Approach
Take the person out of the loop. Put a survey-grade scanner on a mobile robot, let lidar SLAM build the map while it drives, let a frontier-based explorer decide where to go next, and trigger high-resolution scans at waypoints placed on the map the robot just made.
What I built
- A modified SUMMIT-XL platform carrying an Ouster OS1 for live 3D SLAM and a Leica BLK360 for high-resolution capture, driven from ROS on an onboard Intel NUC.
- LIO-SAM for lidar-inertial odometry with loop closure, chosen after benchmarking against LeGO-LOAM and FAST-LIO2 in simulation.
- Frontier-based autonomous exploration, after graph-based planning proved too heavy for the hardware once it met a live site.
- A relational store with one table per ROS topic, hybrid ROSbag and live telemetry capture, and low-latency WebSocket streaming over rosbridge for remote monitoring.
Results
- Three field experiments on an active 500 m² site: a teleoperated baseline, fully autonomous exploration, and waypoint-driven BLK360 scanning.
- The autonomous run covered every critical area and produced a 3D map comparable to the teleoperated baseline, plus the 2D map used to plan the scan waypoints.
- The robot reached every waypoint and completed each scan without human intervention.
- Published at the Creative Construction Conference 2025 as first author. Site access provided by ALEC under the Mubadala-NYUAD Collaborative Project.