Conveyor-Driven Robotic Sorting: Time-Base Regulation, Image-Age Compensation, and Cross-Layer Evaluation Data
Description
This dataset contains the experimental records, supplementary materials, analysis outputs, audit files, post-processing scripts, and provenance resources associated with the study “Traceable Conveyor Time-Base Regulation and Causality-Constrained Image-Age Compensation for Conveyor-Driven Robotic Sorting.” The study treats conveyor speed as the physical time base linking image acquisition to future robotic picking. It evaluates encoder-feedback STEP/DIR regulation with feedforward, saturation handling, anti-windup, and rate limiting, together with a causality-constrained short-window image-age correction. The evidence is organized across control, timing, reference-coordinate reconstruction, scheduler-side pick-window, and task-outcome layers. The formal dataset covers 45 physical trials arranged as G0–G4 × D1tim–D3tim × R1–R3, with a fixed membership of 50 unique tasks per trial. The package also contains separate D0tim implementation-variant records, a representative 30-min G4-D2tim run, Supplementary Fig. S1, Supplementary Tables S1–S15, machine-readable source tables, denominator and causality audits, package manifests, and SHA-256 checksums. Under the prespecified complete-case screen, all 12 closed-loop conditions passed, with nine classified as Stable and three as Usable. For G4, timing-hit rates ranged from 97.33% to 100%; contact was confirmed for all 450 formal tasks, and no task expired. Same-record replay showed that nominal-speed, moving-average, and causal corrections reduced reference-coordinate residuals relative to no correction, but the trial-level intervals did not establish superiority of the causal rule over the simpler alternatives. In the separate 30-min run, contact was confirmed for 239 of 241 timestamped tasks with no expirations. The package supports verification and regeneration of the reported post-processing, task-membership audit, timing-denominator closure, reference-coordinate reconstruction, supplementary tables, and figures from the archived records. Raw logs are treated as primary evidence and derived summaries as secondary cross-checks. The data characterize relative operating margin within the tested protocol; they do not establish absolute host–embedded clock synchronization, independent per-task physical pick-position accuracy, global controller optimality, broad PCB-type generalization, or an empirical robustness boundary.
Files
Steps to reproduce
Download and extract the dataset while preserving the original directory structure. Begin with 00_README_and_Package_Manifest/README.md, which defines the package structure, evidence rules, interpretation boundaries, and inclusion and exclusion decisions. Verify file integrity using SHA256SUMS.txt, FILE_CHECKSUMS_SHA256.csv, and the corresponding package manifests before analysis. For the formal disturbance analysis, use the corrected source bundles in 02_Formal_Trial_Data. The formal matrix contains 45 physical trials defined as G0–G4 × D1tim–D3tim × R1–R3. Each trial uses the fixed 50-task membership recorded in formal_task_membership.csv. The primary membership source is pick_window_test.csv; the documented G0_D3tim_R2 exception uses task_log.csv. Review 06_Postprocessing_Scripts/README_EXECUTION.md before executing the packaged Python scripts. Some scripts retain historical path assumptions, so local input and output paths must be updated without changing the fixed evidence rules or task-membership definitions. Run the relevant packaged scripts to regenerate the supplementary tables, figures, denominator audits, and cross-layer summaries. Compare the regenerated outputs with the reader-facing materials in 01_Supplementary_Materials and the manuscript evidence crosswalk in 05_Manuscript_Source_Tables_and_Figures. Treat the D0tim implementation variants in 03_D0tim_Implementation_Variants and the representative 30-min G4-D2tim records in 04_Audit_and_Auxiliary_Validation as separate auxiliary analyses; they are not part of the formal 45-trial matrix. The package supports reproduction and verification of the reported post-processing, task-membership audit, timing-denominator closure, coordinate-reconstruction analysis, and supplementary outputs from the archived records. It does not recreate the original physical data-acquisition process, establish absolute host–embedded clock synchronization, or claim historical executable identity for the post-experiment M4/M5 reference implementation.