Online Control for Biped Robot with Incremental Learning Mechanism

In this paper, we develop a new online walking controller for biped robots, which integrates neural-network estimator and an incremental learning mechanism to improve the control performance in dynamic environment. With aid of iteration algorithm updating, some newly incoming data can be used straightforwardly update into original well-trained model, order avoid time-consuming retraining procedure. On other hand, how maintain zero-moment-point stability counteract effect yaw moment simultaneously is also key technical problem addressed. To end, interval type-2 fuzzy weight identifier developed, assigns each sample deal with imbalanced distribution training data. The effectiveness proposed scheme has been verified through full-dynamics simulation practical robot experiment.