基於乘坐舒適度量測與深度學習路面辨識應用於智慧型輪椅控制器

Abstract

輪椅乘坐者常為老年人或身障人士，當輪椅行駛於戶外環境，多變的路面型態可能導致乘坐者的不舒適感受，本論文針對路面情況以及當前乘坐舒適度提出智慧輪椅控制器。針對輪椅所行駛的室內外常見環境，以深度學習的技術訓練路面辨識模組，在輪椅行駛的同時，可藉由此模組判斷當前的路面型態。同時，為了顧及乘坐者的乘坐舒適度，採用ISO 2631-1評估舒適度的辦法，以三軸加速度計算加權加速度方均根，並透過大量乘坐測試的問卷填寫訂定輪椅乘坐舒適度標準，藉此評斷行駛的舒適程度。 在控制器的設計上，使用適應性類神經模糊推論系統，依照乘坐者下達的速度命令和當前的乘坐舒適度做為控制器輸入，結合當前路面型態對控制器輸出速度進行些微調整，並加上Q-learning來讓行駛速度能夠更加針對舒適度來即時調整。實驗結果顯示，當輪椅行駛於非常不平坦的路面時，亦能夠保持84.24%的良好乘坐感受。此外，加入路面辨識的控制器相較於未加入路面辨識的控制器，在加速過程中增加了25%的良好乘坐感受，故本論文所提出之智慧輪椅控制器能夠有效的分辨出所在路面型態，並根據所量測到的舒適程度來調整行駛速度，實現基於乘坐舒適度量測與深度學習路面辨識應用於智慧型輪椅。

When the wheelchair is riding in the outdoor environment, the varied types of pavement materials may make people on the wheelchair feel uncomfortable. This paper proposes an intelligent wheelchair controller with pavement recognition and the riding comfort estimation. The deep learning method is used to train a pavement recognition model which can distinguish the current pavement type during the riding process. Meanwhile, it’s necessary to ensure the riding comfort judged by a wheelchair comfort standard defined by the weighted root-mean-square acceleration in ISO 2631-1 and the riding experiment questionnaires. A Q-Learning based Adaptive Network-based Fuzzy Interface System (ANFIS) controller uses the speed command received from the user, the riding comfort index, and the pavement type to adjust the output speed command for the wheelchair. Over 84.24% of the experiment surveys reported good riding comfort from the harshest riding environment. At the same time, the excellent riding comfort feedback percentage from using the controller with the pavement recognition is increased by 25% compared to the percentage from using the controller without the pavement recognition. The intelligent wheelchair is able to effectively distinguish the pavement type and adjust the speed command according to the user requirement of the riding comfort.