Gait abnormalities introduce undesired patterns that limit stability, efficiency, and finally, walker independence. Most of the current algorithms for the identification of gait events using kinematic and inertial data have obtained high performance in healthy subjects. However, most of them showed limited performance when tested with impaired subjects. We hypothesize that to improve gait event detection one must take into consideration the differences between the dynamics of each phase. In this paper, we developed and evaluated a novel methodology for adaptation of a set of parameters for gait event detection using mechanical perturbations. Our proposal employs an ensemble-based procedure for the characterization of the ankle dynamics. We based our proposal on a hybrid model of the human dynamics during gait where the parameters of the dynamics model are abruptly changed according to a Markov chain. In this document, we described the adaptive algorithm and presented preliminary results on a gait simulator. We demonstrated the ability of the algorithm to adapt the parameters according to the changes in human walking.