Searches for pairs of Higgs bosons will be, in all likelihood, the best tools to precisely measure the Higgs boson self-coupling $\lambda_{hhh}$ in future colliders. We study various strategies for the $hh \to b\bar{b}b\bar{b}$ search in the HL-LHC era with focus on constraining $\lambda_{hhh}$. We implement a machine-learning-based approach to separate signal and background and apply recent advances in machine learning interpretability, compare the traditional 4 $b$-jet reconstruction to final states with 1 or 2 large-radius jets, and test scenarios with different top-quark Yukawa couplings, among other factors.