Integrity review: 'Distributed Optimization Algorithm for Multi-Robot Formation with Virtual Reference Center' (Huang et al., IEEE/CAA J. Autom. Sinica, 2022)

Academic fraud report · Geng Detector

Summary

Verdict: Suspicious (yellow), with one orange concern. The paper is a 3-page note extending prior work by the same corresponding author (Qingshan Liu) via a 'virtual reference center' wrapping the Hungarian algorithm, with proofs abbreviated due to page limits. Five findings are reported: (1) ambiguous initialization in Algorithm 1 listing both q and q0 together, suggesting careless editing; (2) an unusually large claimed speed-up over reference [11] (300 outer loops down to 1) reported without Monte Carlo robustness tests, raising cherry-picking risk; (3) substantial reuse of lemmas from references [10] and [11] with skipped proofs; (4) PDF text extraction revealed formula corruption/placeholder characters, limiting verification of derivations; (5) only simulation figures with no hardware validation and no usable raw data. No evidence of outright fabrication was found; concerns are about rigor, reproducibility, and incremental novelty rather than confirmed misconduct.

Verdict

Suspicious (🟡 overall, with one 🟠 concern). No direct evidence of data fabrication, but the paper exhibits weak methodological rigor, heavy self-referential reuse, abbreviated proofs, and a speculative performance claim that together warrant further verification.

Key findings

Algorithm 1 initializes with Initialization: p; q; h; q0, mixing an iteration variable (q) and an initial value (q0) on the same line — semantically ambiguous and suggestive of careless copy-and-edit. (🟡)
Reported vs reference [11]: "the proposed algorithm gets the optimal formation after 10 projection iterations and 1 task assignment with 1 outer loop, but the algorithm in [11] needs 20 projection iterations and 1 task assignment with 300 outer loop." A 300× reduction in outer loops, shown in only two simple simulation figures without Monte Carlo or varied initial conditions, is highly atypical and consistent with favorable parameter selection. (🟠)
Incremental, "re-skinned" novelty: the work rests on references [10] and [11]; the corresponding author Qingshan Liu is also an author of [11]. Lemma 2 and Lemma 3 appear to be carried over from [10], while proofs of Lemma 4 and Theorem 1 are omitted with a "due to the page limit" remark. (🟡)
Text extraction produced malformed formula strings such as minf(q; )= and s:t: 8>>>>><>>>>>: with replacement characters, obstructing independent textual verification of derivations. (🟡)
Only simulation evidence (a 48-point trajectory plot and a smooth objective descent curve); no physical robot experiments and insufficient pixel-level data for forensic image analysis. (🟡)

Evidence highlights

DOI: 10.1109/JAS.2022.105473
Algorithmic ambiguity quote: Initialization: p; q; h; q0 (Algorithm 1, Page 733).
Performance contrast quote: "the proposed algorithm gets the optimal formation after 10 projection iterations and 1 task assignment with 1 outer loop, but the algorithm in [11] needs 20 projection iterations and 1 task assignment with 300 outer loop." (Page 733).
Author overlap with reference [11]: Qingshan Liu appears in both this paper and [11] (Q. Liu and M. Wang, 2021).
Proofs of Lemma 4 and Theorem 1 abbreviated "due to the page limit"; Lemma 2 and Lemma 3 traceable to reference [10].
Only 2 simple simulation figures (Figure 1 trajectory, Figure 2 convergence) across the 3-page paper; no raw dataset or code linked.

Notes

All quantitative and verbatim claims are taken from the supplied review packet; the original PDF was not independently re-mined.
Formula corruption is reported as an extraction artifact; it may reflect non-standard LaTeX or image-embedded equations rather than authoring misconduct.
The 300× speed-up claim could be legitimate under specific tuning; absence of statistical robustness testing is the issue, not provable falsification.
Suggested follow-up: compare derivations against references [10] and [11]; request Monte Carlo simulation code and raw data from the authors; journal should consider tightening acceptance criteria for short simulation-only notes.