Evolutionary Biparty Multiobjective UAV Path Planning: Problems and Empirical Comparisons

The biparty modeling of UAV path planning is well-motivated for urban operations involving separate efficiency and safety departments. The mathematical formulation clearly separates objectives between the efficiency DM ($F_{eff}$) and safety DM ($F_{safe}$), with the former minimizing path length, energy consumption, height variation, and hover point distance while the latter minimizes fatality risk, property risk, and noise pollution. The use of meanHV (averaging hypervolume across decision-makers) provides a balanced evaluation metric for multiparty solutions, and the paper provides detailed parameter settings for reproduction.

The novelty claim of being "the first time" is overstated since the paper applies existing multiparty frameworks rather than inventing new theory. The comparison is limited to only two multiparty algorithms (OptMPNDS and OptMPNDS2) and omits recent decomposition-based MPMOEAs or other clonal selection variants. Table III shows performance differences without statistical significance tests (p-values or confidence intervals), making it impossible to judge if BPAIMA's meanHV of 0.1852 versus BPHEIA's 0.1685 in Case 1 represents a statistically significant improvement. The complexity analysis in Section V-C claims equivalence with NSGA-II by assuming $K \leq m$, but MPNDS2 involves $K+1$ nondominated sorting passes which may incur practical overhead not captured by asymptotic notation.

The evidence demonstrates that standard NSGA-II fails to find common Pareto solutions in BPMOPs, which aligns with expectations given its single-DM design. However, the comparison between the proposed BP variants and OptMPNDS/OptMPNDS2 shows diminishing returns—BPAIMA's improvements over OptMPNDS2 are marginal in Cases 4-6 (0.1979 vs 0.1900, 0.1924 vs 0.1857, 0.1908 vs 0.1841). The paper does not establish whether the adaptive DE strategies in BPAIMA are the decisive factor or simply the immune algorithm framework itself. Furthermore, the paper claims the solution set from single-DM models "is not guaranteed to be Pareto optimal from the perspectives of the efficiency DM and safety DM," but does not quantify how often this theoretical concern manifests in practice.

Experimental parameters are detailed in Section V-A including crossover rates ($CR_1=0.9, CR_2=0.5, CR_3=0.1$), population size 105, and 80,000 evaluations. The code is claimed to be available at a GitHub repository (https://github.com/MiLab-HITSZ/2023ChenMPUAV). However, reproducibility is hampered by the lack of specified random seeds for the urban environment generation—building heights are drawn from lognormal distributions ($\mu=3.04670, \sigma=0.76023$) and population density uses a radial basis model, but the paper does not provide the specific random seeds or instance identifiers needed to regenerate identical test cases. The mission hover point locations (25,30), (34,20), and (40,35) appear arbitrarily selected without justification for these coordinates.