BRICS / 2024-25

Brick Reuse

A robotic inspection and sorting pipeline that determines which reclaimed bricks can return to construction.

Role
Student team member
With
THWS / SWG Schweinfurt
Period
Mar 2024 - Jan 2025
  • YOLOv8-OBB
  • SAM
  • ROS 2

Overview

Construction and demolition work discards large numbers of bricks that remain usable. BRICS, the Brick Reusability and Intelligent Classification System, was a four-student project at THWS run with SWG Schweinfurt. It set out to automate that assessment, from camera detection to robotic sorting.

The supplied demo shows the system on a real UR robot: a wrist-mounted camera detects each brick with an oriented bounding box, the robot grasps it with a custom gripper, and a second camera checks the underside before the brick is sorted.

30
3D-scanned bricks
2,100
images in largest training run
0.96
demo detection confidence

Approach

  1. 01

    Detect from above

    A RealSense camera on the robot's tool head scans the work surface. Oriented-box YOLO models, trained on real images and BlenderProc scenes built from 30 Polycam brick scans with automatic annotations, return each brick's position, rotation, and a first usable call.

  2. 02

    Inspect from below

    Bricks that pass are carried to a single-camera stand that images the bottom face. Both views confirm the grade, with Segment Anything isolating mortar residue and surface damage on each face.

  3. 03

    Grip and sort

    ROS 2 and MoveIt plan every motion under hand-eye calibration, while widened Robotiq fingers with 3D-printed TPU surfaces keep irregular, mortar-covered bricks from slipping. Confirmed bricks are set aside for reuse; the rest are discarded.

Rendered pallet scenes with automatic rotated bounding boxes and usable/unusable masks
Synthetic training data: rotated boxes and class masks are generated automatically with every render.

Outcome

The demonstration pipeline ran on the real robot: the supplied footage shows a brick classified as usable at 0.96 confidence and carried away cleanly. The original five-camera inspection stand was simplified to a single camera without losing classification accuracy, and a feeder-and-conveyor pipeline was designed as the production-stage successor.