Algorithm 1 Online Modified MPIO Teacher for Per-step Weight Selection

Require:  Local state of UAV i, neighbor states, params $P,K_d,K_{max}$ Ensure:  Teacher weight ${\mathbf{w}}_i^{\star }$   1:Initialize pop. $\mathcal{P}={\{{\mathbf{X}}_q,{\mathbf{V}}_q\}}_{q=1}^P$ in ${[0,1]}^2$ and inject previous valid weight.   2:Evaluate costs for all pigeons. Mark infeasible candidates ($j_{3,i}=1$ or $j_{4,i}=1$) to exclude them from Pareto ranking; init elite archive $A\leftarrow \mathsf{\varnothing }$.   3:for  $k=1,\dots ,K_{max}$  do   4:    Pareto-sort the feasible subset of $\mathcal{P}$, compute landmark center ${\mathbf{X}}_{\mathrm{center}}^{\left(k\right)}$ via Equation (18).   5:    Update $A\leftarrow ArchiveUpdate(A,{\mathcal{F}}_1^{\left(k\right)})$ and select global guidance ${\mathbf{X}}_g^{\left(k\right)}$ from ${\mathcal{F}}_1^{\left(k\right)}$.   6:    for each pigeon $q\in \mathcal{P}$ do   7:        Store ${\mathbf{X}}_q^{\mathrm{old}}$.   8:        Update: If q is a leader, update via Equations (19) and (20); else perform hierarchical follower update (Equation (21)) or random walk.   9:        Evaluate & Rollback: Project to bounds and evaluate costs. Mark if infeasible. If dominated by its previous state, roll back to ${\mathbf{X}}_q^{\mathrm{old}}$ and restore costs. 10:    end for 11:    Remove the worst $K_d$ pigeons based on Pareto rank and crowding distance to reduce $\left|\mathcal{P}\right|$. 12:end for 13:Extract the final feasible Pareto front set ${\mathcal{S}}_1$ from $\mathcal{P}$. 14:return ${\mathbf{w}}_i^{\star }\in {\mathcal{S}}_1$ minimizing $j_{2,i}$, else best feasible in A, else default $[0.2,0.8]$.