Belangrijkste concepten
Proposing a semi-supervised method with loss regularization for accurate terrain classification in legged robots.
Statistieken
The accuracy of the proposed method is around 89%.
Proposed method improves by 22% over SVM and 9% over FCN based techniques.
Proposed method even improves by 1.5% over TCN based technique.
Citaten
"The proposed method has contributed twofold, viz. by adopting a stacked LSTM architecture, and by including a new loss regularization approach."
"These developments have made possible their use supplementing the state-of-the-art in emergency scenarios."