1,100 Stone Meshes and Placement Candidates for Robotic Dry Stone Construction

1,100 digitized stone meshes, including quarried boulders and concrete debris, were collected by the autonomous excavator HEAP. The dataset from ETH Zurich includes raw and cleaned 3D meshes, candidate placements for automated wall building, and hand-labeled viability scores. It supports research in robotic excavation and autonomous construction using on-site materials.

Use Cases

• Training or evaluating robotic grasp and placement models based on 3D stone mesh data.
• Developing classifiers for construction viability using hand-labeled candidate placement attributes.
• Simulating dry stone wall construction scenarios using provided point cloud and SDF representations.
• Benchmarking 3D reconstruction and shape analysis algorithms on raw and processed stone geometries.

Strengths

• Contains 1,100 individual stone meshes in both raw and cleaned, closed formats.
• Includes candidate placement data with hand-labeled binary viability scores.
• Provides multiple data representations: meshes, point clouds (.pcd), and signed distance fields (.npy).

Limitations

• Row count and dataset size are unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Ryan Luke Johns, ETH Zurich

Collection Method

Stones were digitized by the autonomous excavator HEAP using LiDAR scanning and Poisson reconstruction.