基於深度學習與感測器融合的高機動性行走機器人助行器

Abstract

老年人口面臨諸多行動困難－即便僅在家中短距移動亦然。各式行動輔具如輪椅、拐杖及助步器等應運而生。這些儀器皆可藉由調整其電子或電機元件增進其功能。添加引擎、編碼器、力感測器、景深攝影機及雷射測距儀之CAIROW(情境感知智慧型行動輔具)即為一例。 本論文目的為透過藉由深度學習為基石之數據分析方法，分析使用者自由移動機器所蒐集之實驗數據(機器被動接收使用者提供之步態、施力及各式情境數據)以增進CAIROW的可操作性。使用CNN(卷積神經網路)及LSTM(長短期記憶) 於不同情境及步態下習得最佳發送引擎資訊。並透過深度學習技術融合異質性高之感測器接收資訊。因此，此項論文全然以數據為基礎，亦即機器於監督式學習下透過與環境互動習得最佳化行為。最後請使用者於接受測試後定性評估機器操作難易度。

Elder population face a lot of challenges when it comes to locomotion activities. Even walking for short distances inside the home might be difficult for them. To address this problem, there are different mobility-aid devices such as: wheelchairs, canes, or walkers. All of these devices can be modified with electronic and electrical components in order to improve their functionality. A good example is CAIROW (Context Awareness Intelligent Robotic Walker) which was equipped with motors, encoders, force sensors, depth camera, and a laser range finder. The objective of this thesis is, then, improve the control and maneuverability of CAIROW through the analysis of data gathered from experiments of the motion of the robot in a passive way. The methodology used to analyze the data is based on Deep Learning techniques. CNN (Convolutional Neural Networks) and LSTM (Long-Short Term Memory) are two major techniques applied to learn the best motor signal under different scenarios and gait requirements. Even though the signals from the sensors are very heterogeneous, the Deep Learning techniques are plausible to fuse them. Therefore, this work is totally data driven where the robot learns the optimal behavior based on the interactions with the environment in a supervised learning approach. Finally, the robot is tested with some users who provide qualitative comments about the comfort of using the robot.