The constant evolution of personal urban driving habits is surely pushing the rapid proliferation of new micromobility solutions, especially in the form of eScooters, which are emerging as a green, comfortable, and inexpensive alternative to classical means of transportation. However, the diffusion of such kind of light electric vehicles is accompanied by an increase in the related accidents, often due to risky driving styles and behaviours non-compliant with the traffic rules. In particular, the majority of the injuries are associated to a travelling speed inadequate to the quality of the road surface covered and to the presence of a second passenger, which makes the vehicle more difficult to be handled, besides being forbidden by the law. In this perspective, this study proposes three safety-oriented data-driven algorithms to ensure the driver's security and to protect the health status of the vehicle, retrieving a control logic that, under specific conditions, opportunely limits the eScooter speed or even induces its shut-down. The novelty of this research is to address issues never tackled for this class of vehicles up to now, relying only on a standard sensor equipment, mainly composed of an Inertial Measurement Unit and a Hall-effect sensor for the measurement of the user's gas request at the throttle. To obtain an estimation of the travelled surface, the Road Quality Index (RQI) has been designed simply referring to signal processing and feature extraction techniques, achieving the goal of recognising the 5 main road scenarios encountered in an urban environment. Concerning the detection of a second onboard passenger, an algorithm based on a Recursive Least Squares with exponentially varying forgetting factor (μ^exp-RLS) has been leveraged to retrieve an estimate of the driver mass. The obtained mass value is then exploited for the development of a Boosted trees-based (BTB) classifier that performs binary classifications between single and double passenger. The ad-hoc designed experimental campaigns proved the effectiveness of all three estimation algorithms, assessing also their optimal generalisation capabilities, which are strongly required in the optic of a future online implementation.
La costante evoluzione delle abitudini di guida personali sta indubbiamente agevolando la rapida diffusione di nuove soluzioni di micromobilità, specialmente nella forma di monopattini elettrici (eScooters), che stanno emergendo come una alternativa green, confortevole ed economica ai classici mezzi di trasporto. Tuttavia, la diffusione di questi nuovi veicoli elettrici leggeri è accompagnata da un netto aumento degli incidenti ad essi legati, spesso dovuti a condotte di guida pericolose e comportamenti non conformi alle regole stradali. In particolare, la maggior parte degli infortuni sono associati a velocità di guida inadeguate alla qualità della superficie stradale percorsa e alla presenza a bordo di un secondo passeggero, che rende più difficile il controllo del veicolo, oltre che ad essere proibito dalla legge. In questa ottica, questo studio propone tre algoritmi data-driven volti a garantire la sicurezza del guidatore e del veicolo stesso, tramite lo sviluppo di una logica di controllo che, sotto determinate condizioni, limita la velocità del monopattino o addirittura ne induce lo spegnimento. L'aspetto innovativo di questa ricerca sta nell'affrontare problemi finora mai trattati per questa classe di veicoli, sfruttando un pacchetto sensoristico standard, composto da una Inertial Measurement Unit ad un sensore ad effetto Hall per la misura della richiesta di gas dell'utente all'acceleratore. Per ottenere una stima della superficie stradale percorsa, è stato definito attraverso semplici tecniche di signal processing ed estrazione di features il Road Quality Index (RQI), un indice capace di riconoscere i 5 principali scenari stradali che possono essere incontrati in un contesto urbano. Per quanto riguarda il rilevamento di un secondo passeggero a bordo, è stato sviluppato un algoritmo per stimare la massa del guidatore utilizzando un Recursive Least Squares con fattore di oblio ad andamento esponenziale (μ^exp-RLS). Il valore di massa ottenuto è stato poi sfruttato per la realizzazione di un classificatore Boosted trees-based (BTB), che ha il compito di eseguire classificazioni binarie tra le classi di singolo e doppio guidatore. Le campagne sperimentali appositamente condotte hanno dimostrato l'efficacia di tutti e tre algoritmi di stima, certificando allo stesso tempo delle ottime capacità di generalizzazione, strettamente richieste in prospettiva di una futura implementazione online sul veicolo.
Safety-oriented real-time algorithms for road quality, driver mass and number of passengers estimation in urban eScooters
Paris, Jacopo
2019/2020
Abstract
The constant evolution of personal urban driving habits is surely pushing the rapid proliferation of new micromobility solutions, especially in the form of eScooters, which are emerging as a green, comfortable, and inexpensive alternative to classical means of transportation. However, the diffusion of such kind of light electric vehicles is accompanied by an increase in the related accidents, often due to risky driving styles and behaviours non-compliant with the traffic rules. In particular, the majority of the injuries are associated to a travelling speed inadequate to the quality of the road surface covered and to the presence of a second passenger, which makes the vehicle more difficult to be handled, besides being forbidden by the law. In this perspective, this study proposes three safety-oriented data-driven algorithms to ensure the driver's security and to protect the health status of the vehicle, retrieving a control logic that, under specific conditions, opportunely limits the eScooter speed or even induces its shut-down. The novelty of this research is to address issues never tackled for this class of vehicles up to now, relying only on a standard sensor equipment, mainly composed of an Inertial Measurement Unit and a Hall-effect sensor for the measurement of the user's gas request at the throttle. To obtain an estimation of the travelled surface, the Road Quality Index (RQI) has been designed simply referring to signal processing and feature extraction techniques, achieving the goal of recognising the 5 main road scenarios encountered in an urban environment. Concerning the detection of a second onboard passenger, an algorithm based on a Recursive Least Squares with exponentially varying forgetting factor (μ^exp-RLS) has been leveraged to retrieve an estimate of the driver mass. The obtained mass value is then exploited for the development of a Boosted trees-based (BTB) classifier that performs binary classifications between single and double passenger. The ad-hoc designed experimental campaigns proved the effectiveness of all three estimation algorithms, assessing also their optimal generalisation capabilities, which are strongly required in the optic of a future online implementation.| File | Dimensione | Formato | |
|---|---|---|---|
|
2020_12_Paris.pdf
non accessibile
Dimensione
29.22 MB
Formato
Adobe PDF
|
29.22 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/170664