"Autonomous Robot For Indoor Enhanced Living (ariel) " by Prabhjot Kaur

Access Type

Open Access Dissertation

Date of Award

January 2024

Degree Type

Dissertation

Degree Name

Ph.D.

Department

Computer Science

First Advisor

Weisong W. Shi

Second Advisor

Zheng Z. Dong

Abstract

The global population is aging rapidly, with the total percentage of older adults (65 years and older) projected to increase from 10\% of the total population in 2022 to 16\% by 2050, according to the World Population Prospectus 2022 issued by the United Nations. For certain parts of the world such as Europe and North America, this translates to 1 in every 4 persons is projected to be 65 years or older by 2060. This trend raises concerns about providing quality long-term care for the older population. Moreover, according to the 2021 survey by the American Association of Retired Persons (AARP), more than three-quarters of adults wish to age in their own homes and communities. This is expected to create an increase in demand for long-term care. Given that the current long-term care system is already struggling with a shortage of caregivers, it is crucial to implement innovative ways to ensure a sustainable long-term care system. We propose integrating autonomous mobile robots into the existing long-term care network to help adults age in place without jeopardizing their safety. Our research exploits ambient sound and speech from indoor environments to develop various applications that can be deployed on an autonomous mobile robot.

Firstly, we introduce ListenBot, a robot audition system that acts as a safety and security guard for indoor environments. The system classifies and localizes ambient sounds to inform of any safety and security threats. The results show that the average error of the SSL in computing the azimuth angle of the sound source is less than 3.3304$^{\circ}$ for a source that is up to 3 meters away from the robot’s position. Secondly, given that falls are one of the deadly threats to the health of older adults, we developed a fall detection system based on audio input. This non-wearable and non-invasive system classifies falls with an accuracy of 0.8673. Thirdly, we augment the fall detection system with the addition of speech recognition for distress detection from speech. Since falls are often accompanied by speech such as screaming or calling for help, we detect speech using a state-of-the-art speech recognition model called Whisper. Lastly, we present a direct speech-to-speech translation application. We demonstrate a novel end-to-end speech translation model on Punjabi to English translation though the model is extensible to any other pair of languages. The model achieves a BLEU score of 3.9829.

Download

Share

COinS