Publicação
Design and evaluation of a cognitive vehicle system: emphasizing user routine learning and interaction
| Resumo: | Driving a vehicle is often a routine activity that involves visiting the same places at roughly the same times on specific days of the week. Hence, personal vehicles typically transport the same occupants along with their specific items. Therefore, the development of cognitive/intelligent vehicles capable of learning and adapting to their occupants’ routines can assist drivers autonomously during their daily trips. This dissertation aims to develop an integrated system that allows extraction of stop locations from the user’s mobility information and learning of their respective spatio-temporal routines. The system comprises a User Interface/Authentication module, User Stop Location Extraction module, and a User Routine Learning Cognitive module. The User Stop Location Extraction module consists in a machine learning approach, incorporating clustering and artificial neural networks methods to identify and classify stop locations in a given user mobility dataset. This module interacts and supports an already existent User Routine Learning Cognitive module, based on Dynamic Neural Fields (DNFs), which is responsible for the implicit learning of various destinations and their associated time characteristics. Additionally, the Interface/Authentication module involves the development of a basic interface to aid in visualizing the resulting stop locations, provide basic User-Vehicle interaction with the system, and enable user authentication. Experiments conducted with real data regarding the routines of various vehicles have validated the newly developed modules and respective integration with cognitive module. The results notably highlighted the Stop Location Extraction module's particularly satisfactory performance in accurately identifying and classifying stop locations. This module also allows visualization of stop locations thus enhancing the interpretability of predictions generated by the cognitive module when compared to a prior Stop Location Extraction approach based on DNFs. However, despite the improved performance of the new Stop Location Extraction module, it did not necessarily translate into enhanced prediction accuracy by the cognitive module. The prediction errors occur in cases of exceptional, one-time deviations from routine behavior. In essence, this dissertation has contributed to further advance the research within Project BEN, specifically aiming to equip BEN vehicles with cognitive capabilities. |
|---|---|
| Autores principais: | Dias, Pedro Henrique Sampaio |
| Assunto: | Human-vehicle interaction Intelligent vehicles Learning user habits Spatial clustering Agrupamento espacial Aprendizagem das rotinas do utilizador Interação humano-veículo Veículos inteligentes |
| Ano: | 2024 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade do Minho |
| Idioma: | inglês |
| Origem: | RepositóriUM - Universidade do Minho |
| Resumo: | Driving a vehicle is often a routine activity that involves visiting the same places at roughly the same times on specific days of the week. Hence, personal vehicles typically transport the same occupants along with their specific items. Therefore, the development of cognitive/intelligent vehicles capable of learning and adapting to their occupants’ routines can assist drivers autonomously during their daily trips. This dissertation aims to develop an integrated system that allows extraction of stop locations from the user’s mobility information and learning of their respective spatio-temporal routines. The system comprises a User Interface/Authentication module, User Stop Location Extraction module, and a User Routine Learning Cognitive module. The User Stop Location Extraction module consists in a machine learning approach, incorporating clustering and artificial neural networks methods to identify and classify stop locations in a given user mobility dataset. This module interacts and supports an already existent User Routine Learning Cognitive module, based on Dynamic Neural Fields (DNFs), which is responsible for the implicit learning of various destinations and their associated time characteristics. Additionally, the Interface/Authentication module involves the development of a basic interface to aid in visualizing the resulting stop locations, provide basic User-Vehicle interaction with the system, and enable user authentication. Experiments conducted with real data regarding the routines of various vehicles have validated the newly developed modules and respective integration with cognitive module. The results notably highlighted the Stop Location Extraction module's particularly satisfactory performance in accurately identifying and classifying stop locations. This module also allows visualization of stop locations thus enhancing the interpretability of predictions generated by the cognitive module when compared to a prior Stop Location Extraction approach based on DNFs. However, despite the improved performance of the new Stop Location Extraction module, it did not necessarily translate into enhanced prediction accuracy by the cognitive module. The prediction errors occur in cases of exceptional, one-time deviations from routine behavior. In essence, this dissertation has contributed to further advance the research within Project BEN, specifically aiming to equip BEN vehicles with cognitive capabilities. |
|---|