BRAC-ACT: Supporting Data for Timber Dougong Assembly Experiments

Supporting data and scripts accompany a study on BRAC-ACT, a structured imitation-learning framework for long-horizon traditional timber dougong assembly. The deposit contains processed materials for verifying reported analyses, including task definitions, configuration files, training parameters, logs, and evaluation tables. The dataset was published by an anonymous researcher on figshare in May 2026.

Use Cases

• Reproduce experimental results based on provided training logs and evaluation tables.
• Configure robot and camera setups for similar assembly tasks based on provided configuration files.
• Analyze subtask performance in a chained evaluation workflow based on the described experimental structure.

Strengths

• Includes a detailed experimental workflow description covering data collection, screening, training, and evaluation.
• Provides specific file types for verification, such as task definitions, training parameters, and checkpoint summaries.
• Released under a permissive CC-BY-4.0 license.

Limitations

• The core raw multi-camera demonstration dataset is not included; only processed supporting data is provided.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Anonymous Researcher via figshare

Collection Method

Generated from a subtask-wise experimental workflow involving data collection, manual screening, and model training.

Freshness

Last updated 2026-05-01 02:04:01