Bio-inspired hierarchical design combined with advanced computational modelling provides an effective route for developing next-generation lightweight crashworthy structures. This study presents a computational design framework for hierarchical thin-walled tubes, progressing from tree-inspired hexagonal configurations to self-similar nested tubes with controllable internal scale, thickness redistribution, oblique collapse assessment, and surrogate-assisted optimisation. High-fidelity nonlinear finite element simulations, validated against axial compression experiments, are used to investigate crushing responses, deformation modes, plastic strain evolution, and energy dissipation mechanisms. The results show that hierarchical refinement improves crashworthiness only when internal members deform cooperatively, with the second-order self-similar nested tube (H-SNT-2) providing the best balance between hierarchy activation and deformation efficiency under equal-mass axial crushing. Axial thickness gradients tune folding initiation and peak response, while radial gradients redistribute plastic dissipation among structural layers. Oblique loading further reveals a transition from mixed collapse to Euler-buckling-dominated deformation, highlighting the need to delay global bending and maintain progressive folding. Finally, a CatBoost surrogate model coupled with NSGA-II optimisation identifies balanced low-IPCF and high-SEA designs. Overall, the study demonstrates that crashworthiness is governed by coordinated control of topology, internal scale, thickness distribution, and collapse mode, rather than geometric complexity alone.