Provided dataset is related to the following publication:

Paper:
A Feedback-Driven Framework for Adaptive Multi-Robot Construction

Authors:
Arash Adel, Daniel Ruan, Wesley McGee, and Salma Mozaffari

Institutions:
- Princeton University
- University of Michigan

Journal:
Automation in Construction

Date:
2023-2024

Funding:
This research was supported by the National Science Foundation (NSF,
Award No. 2128623) and the Taubman College of Architecture and Urban Planning at
the University of Michigan (U-M).

Description:
The deposit contains data related to the adaptive multi-robotic assembly of
timber structures. This data was collected across two experiments; the first
experiment utilizes a pose-based adaptive fabrication method applied to a
nail-laminated timber module, and the second experiment utilizes a topology-
based adaptive fabrication method applied to a spatial wall frame module.

The first experiment data contains the pose measurements for each element in the
assembly structure and overall point cloud scans of the final as-built
assemblies. Pose measurements were derived from a defined set of profile scans
using a profile scanner attached to the end of an industrial robot arm, using
the methods detailed in the corresponding paper. This scanner has a profile
resolution of 0.150 mm, a depth resolution of 0.019 mm, and a linearity of
±0.01% of the measurement range. These pose measurements were used as inputs to
the adaptive control loop, and also served as measures of deviation when
compared to a benchmark case assembly with no adaptation.

The point cloud scans were likewise obtained using the profile scanner by
controlling the industrial robot to perform a linear sweep across the final
structure, capturing a profile of points every 2 mm. Since the field of view of
the sensor was insufficient to scan the entire structure in a single pass,
multiple passes were made with a slight overlap, and these point cloud segments
were stitched together using the calibrated end-effector position of the
industrial robot. The point clouds were manually filtered to remove background
elements, such as the assembly platform and clamping elements. These point
clouds were utilized to compute surface deviations of the adaptive and benchmark
assembly cases against an idealized digital model. The deviation data
corresponds to the order of points in their corresponding point cloud files.

The second experiment did not utilize pose measurements, and, therefore, the
database only includes the point cloud scan and deviation calculations of the
as-built wall module. For more details, please see the corresponding paper.

********************************************************************************

Database contains all the data presented in Section 4 of the publication. 

01_Experiment1/				NLT panel experiment (Section 4.1)
├─ 01_PoseDeviations/			Summarized in Table 1 (tracking error)
│  ├─ AdaptivePose.csv
│  ├─ BenchmarkPose.csv
├─ 02_PointCloudScans/			Plotted and visualized in Fig. 12
│  ├─ 01_FilteredScans/
│  │  ├─ AdaptiveScan.ply		557,146 points
│  │  ├─ BenchmarkScan.ply		600,017 points
│  ├─ 02_PointCloudDeviations/
│  │  ├─ AdaptiveDeviation.csv		Corresponds to AdaptiveScan.ply, in mm
│  │  ├─ BenchmarkDeviation.csv		Corresponds to BenchmarkScan.ply, in mm
02_Experiment2/				Wall module experiment (Section 4.3)
├─ 01_PointCloudScans/			Plotted and visualized in Fig. 13
│  ├─ 01_FilteredScans/
│  │  ├─ WallModuleScan.ply		1,816,388 points
│  ├─ 02_PointCloudDeviations/
│  │  ├─ WallModuleDeviation.csv	Corresponds to WallModuleScan.ply, in mm