The project was developed during the Leaders of Digital Transformation hackathon, addressing a task from the Moscow Department of Transport. The goal was to create an algorithm for obstacle detection in subway tunnels for autonomous trains.
The system utilizes data from a Hesai 128 3D lidar to identify obstacles within the track gauge at distances up to 150+ meters in complete darkness. The primary requirement was to maintain minimal false-positive rates.
The developer implemented a digital twin of the infrastructure based on spatial memory rather than traditional frame-by-frame polynomial calculations. The pipeline includes odometry, a spatial memory concept, and a 3D cockpit interface built on Three.js.
The project did not reach the top 10 due to technical issues with the automated judging script, specifically related to data output structure and minor synchronization errors in extrapolation. The core of the project is now available under an MIT open-source license.
