Robust Embodied Self-Identification of Morphology in Damaged Multi-Legged Robots

To further validate the algorithm's convergence and robustness, we repeated the damage identification process 10 times for each test scenario. As an example, the best objective function value per generation for the Legs 4 and 5 missing scenario is shown in Figure 1. Across the 10 identification runs, the resulting morphology was either identical or differed by a single link within the identified damaged legs, a discrepancy discussed before. The most frequently identified morphology is reported in Table III, representing the most probable morphological configuration based on the algorithm's convergence behavior. As shown in Figure 1, the algorithm converged to three distinct morphologies over 10 attempts.