Evaluating the Safety and Performance of Electric Micro-Mobility Vehicles Comparing E-bike, E-scooter and Segway based on Objective and Subjective Data from a Field Experiment Master’s thesis in Systems, Control and Mechatronics LUCAS BILLSTEIN CHRISTOFFER SVERNLÖV DEPARTMENT OF MECHANICS AND MARITIME SCIENCES CHALMERS UNIVERSITY OF TECHNOLOGY Gothenburg, Sweden 2021 www.chalmers.se www.chalmers.se Master’s thesis 2021:36 Evaluating the Safety and Performance of Electric Micro-Mobility Vehicles Comparing E-bike, E-scooter and Segway based on Objective and Subjective Data from a Field Experiment LUCAS BILLSTEIN CHRISTOFFER SVERNLÖV Department of Mechanics and Maritime Sciences Division of Vehicle Safety Unit of Crash Analysis and Prevention Chalmers University of Technology Gothenburg, Sweden 2021 Evaluating the Safety and Performance of Electric Micro-Mobility Vehicles Comparing E-bike, E-scooter and Segway based on Objective and Subjective Data from a Field Experiment LUCAS BILLSTEIN CHRISTOFFER SVERNLÖV © LUCAS BILLSTEIN, 2021. © CHRISTOFFER SVERNLÖV, 2021. Supervisor: Alexander Rasch, Department of Mechanics and Maritime Sciences Supervisor: Christian-Nils Åkerberg Boda, Autoliv Examiner: Marco Dozza, Department of Mechanics and Maritime Sciences Master’s Thesis 2021 Department of Mechanics and Maritime Sciences Division of Vehicle Safety Unit of Crash Analysis and Prevention Chalmers University of Technology SE-412 96 Gothenburg Telephone +46 31 772 1000 Cover: A picture of the e-bike, e-scooter and Segway used in the study. Typeset in LATEX, template by Magnus Gustaver Department of Mechanics and Maritime Sciences Gothenburg, Sweden 2021 iv Evaluating the Safety and Performance of Electric Micro-Mobility Vehicles Comparing E-bike, E-scooter and Segway based on Objective and Subjective Data from a Field Experiment LUCAS BILLSTEIN CHRISTOFFER SVERNLÖV Department of Mechanics and Maritime Sciences Chalmers University of Technology Abstract The rapid increase in popularity of electric micro-mobility vehicles has led to an increasing amount of injuries. A large proportion of injuries are related to the ve- hicles themselves, therefore, a thorough investigation of their safety is needed. The purpose of this thesis was to collect conclusive evidence for or against the safety level of e-bikes, e-scooters and Segways in terms of stability, maneuverability and rider comfort. A regular bike was used as well to compare with. Rider kinematics data were recorded by sensors mounted on each of the vehicles together with a stationary LIDAR sensor. Thirty-four voluntary participants performed four different tasks with each of the vehicles. Afterwards, they filled in a questionnaire regarding the experienced safety and performance of the vehicles. A set of performance indicators were studied to be able to compare the vehicles and establish the safety of each vehicle. Results suggest that the e-bike and the e-scooter performed very well in terms of rider comfort and stability. The e-scooter also had good maneuverability, but lacked in safety due to bad braking performance. The least safe vehicle was the Segway which was rated the least safe by the participants and performed the worst according to the performance indicators in high speed scenarios. There were however scenarios at low speed, with the Segway that had comparable results with the other vehicles. Future studies could include other micro-mobility vehicles or other tasks. Exper- iments in a more naturalistic settings could provide valuable data to extend the results presented in this thesis. The results from this thesis could be used to help improve the design of electric micro-mobility vehicles and contribute to guidelines for infrastructure design and policy making. Keywords: safety, performance, micro-mobility, e-scooter, e-bike, Segway v Acknowledgements Thank you to our supervisor Alexander Rasch for supporting our work with thought- ful insights and recommendations, and for always gladly answering to questions. Your support and positive comments helped us get through stressful periods of this thesis. Thank you to our supervisor from Autoliv, Christian-Nils Åkerberg Boda for all the help and discussion about technical details such as the orientation estimate. Your humor and positive attitude was always welcome. Thank you Christian-Nils for al- ways thinking one step further and challenging us throughout the whole thesis. Thank you to Marco Dozza, our examiner, for showing up on our meetings with great insights and ideas. Thank you so much for making this master’s thesis possi- ble. Thank you to all three, Alexander, Christian-Nils and Marco for being very easy to work with. Thank you to Alessio Violin for giving an introduction to the instrumentation. Thank you to our opponents, Felicia Österberg and Paula Ek, for their great dis- cussion after our presentation and their detailed feedback on our report. Thank you to SAFER for advertising our poster that was made to gather partici- pants. Thank you to all the participants that participated in the field experiment. Without all 34 of you, this thesis would not have turned out the way it did. Thank you to the kind employees at Autoliv which listened to our "mid-term" pre- sentation and gave great feedback. We would also like to thank Trafikverket for sponsoring this thesis through the Skylt- fonden Project "Characterizing and classifying new e-vehicles for personal mobility (TV 2019/21327)". Lucas Billstein, Gothenburg, June 2021 Christoffer Svernlöv, Gothenburg, June 2021 vii Contents List of Figures xiii List of Tables xv 1 Introduction 1 1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Previous Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 1.3 Research Question . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.4 Purpose . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.5 Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 2 Methodology 5 2.1 Vehicle Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 2.1.1 E-bike . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 2.1.2 E-scooter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 2.1.3 Segway . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 2.2 Data Acquisition System . . . . . . . . . . . . . . . . . . . . . . . . . 7 2.2.1 Data Logger . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.2.2 Inertial Measurement Unit . . . . . . . . . . . . . . . . . . . . 9 2.2.3 Steering Angle Sensor . . . . . . . . . . . . . . . . . . . . . . 10 2.2.4 LIDAR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 2.3 Participant Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 2.4 Experimental Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . 13 2.4.1 Description of Tasks . . . . . . . . . . . . . . . . . . . . . . . 13 2.4.2 Set Up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 2.4.3 Main Experiment . . . . . . . . . . . . . . . . . . . . . . . . . 16 2.4.4 Questionnaire . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 2.5 Performance Indicators . . . . . . . . . . . . . . . . . . . . . . . . . . 17 2.5.1 Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 2.5.2 Braking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 2.5.3 Acceleration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 2.5.4 Steering and Leaning . . . . . . . . . . . . . . . . . . . . . . . 21 2.5.4.1 E-bike, Bike and E-scooter . . . . . . . . . . . . . . . 21 2.5.4.2 Segway . . . . . . . . . . . . . . . . . . . . . . . . . 22 2.5.5 Box Plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 2.6 Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 ix Contents 2.6.1 Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 2.6.2 Post Processing . . . . . . . . . . . . . . . . . . . . . . . . . . 24 2.6.2.1 Syncing the Times . . . . . . . . . . . . . . . . . . . 24 2.6.2.2 Filtering . . . . . . . . . . . . . . . . . . . . . . . . . 24 2.6.2.3 Segmentation . . . . . . . . . . . . . . . . . . . . . . 25 2.6.2.4 Steering Angle Correction . . . . . . . . . . . . . . . 26 2.6.2.5 Orientation Estimation . . . . . . . . . . . . . . . . . 26 2.6.2.6 LIDAR Data Processing . . . . . . . . . . . . . . . . 26 2.6.2.7 Rauch-Tung-Striebel Smoother . . . . . . . . . . . . 27 3 Results 29 3.1 Data Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 3.2 Performance Indicators . . . . . . . . . . . . . . . . . . . . . . . . . . 31 3.2.1 Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 3.2.2 Braking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 3.2.3 Accelerating . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37 3.2.4 Steering and Leaning . . . . . . . . . . . . . . . . . . . . . . . 39 3.2.4.1 E-bike, Bike and E-scooter . . . . . . . . . . . . . . . 39 3.2.4.2 Segway . . . . . . . . . . . . . . . . . . . . . . . . . 41 3.3 Questionnaire . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 3.3.1 Mounting and Dismounting . . . . . . . . . . . . . . . . . . . 43 3.3.2 Accelerating from Standing Still . . . . . . . . . . . . . . . . . 43 3.3.3 Turning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 3.3.4 Maintaining Low Speed . . . . . . . . . . . . . . . . . . . . . 44 3.3.5 Maintaining High Speed . . . . . . . . . . . . . . . . . . . . . 44 3.3.6 Keeping Balance . . . . . . . . . . . . . . . . . . . . . . . . . 44 3.3.7 Braking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 3.3.8 Steering at Low Speed . . . . . . . . . . . . . . . . . . . . . . 45 3.3.9 Steering at High Speed . . . . . . . . . . . . . . . . . . . . . . 45 3.3.10 Overall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 4 Discussion 47 4.1 Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 4.1.1 Speed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 4.1.2 Braking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 4.1.3 Acceleration . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48 4.1.4 Steering and Leaning . . . . . . . . . . . . . . . . . . . . . . . 49 4.1.5 Mounting and Dismounting . . . . . . . . . . . . . . . . . . . 50 4.2 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50 4.2.1 Data Sets Availability . . . . . . . . . . . . . . . . . . . . . . . 50 4.2.2 Potentiometer . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 4.2.3 Measurement Uncertainty . . . . . . . . . . . . . . . . . . . . 51 4.2.4 Biases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 4.3 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 5 Conclusion 53 x Contents A Data Set Availability I B Questionnaire III C Task Description Scripts IX C.1 Script for E-bike, Bike and E-scooter . . . . . . . . . . . . . . . . . . IX C.2 Script for Segway . . . . . . . . . . . . . . . . . . . . . . . . . . . . . X xi Contents xii List of Figures 2.1 Monark Karin 3-VXL electric bike. . . . . . . . . . . . . . . . . . . . 5 2.2 Ninebot Kickscooter ES2. . . . . . . . . . . . . . . . . . . . . . . . . 6 2.3 Segway Ninebot S. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 2.4 Vehicle instrumentation. . . . . . . . . . . . . . . . . . . . . . . . . . 8 2.5 Data logger which is mounted on the e-bike. On top of the data logger is an IMU mounted. . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.6 Steering angle sensor mounted on the steering stem of the e-bike and e-scooter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 2.7 IMU mounted on the top of the steering stick of the Segway. . . . . . 11 2.8 LIDAR mounted on a tripod. . . . . . . . . . . . . . . . . . . . . . . 12 2.9 The first task which is to study the behavior of constant high speed, comfortable braking and comfortable acceleration. . . . . . . . . . . . 13 2.10 The second task which studies the behavior of constant high speed, harsh braking and harsh acceleration. . . . . . . . . . . . . . . . . . . 14 2.11 The third task which studied the behavior of a harsh acceleration and an unexpected harsh brake. . . . . . . . . . . . . . . . . . . . . . . . 14 2.12 The fourth task which studies the behavior during consecutive turns. 15 2.13 Illustration of the test area. . . . . . . . . . . . . . . . . . . . . . . . 15 2.14 Photo of the middle part of the test area. . . . . . . . . . . . . . . . . 16 2.15 Description of a box chart. . . . . . . . . . . . . . . . . . . . . . . . . 23 2.16 Process description for post processing. . . . . . . . . . . . . . . . . . 24 2.17 Example of the segmentation for the braking tasks. . . . . . . . . . . 25 2.18 An example of the segmentation for the slalom task. The origin is the position of the LIDAR. . . . . . . . . . . . . . . . . . . . . . . . . 26 2.19 A speed estimate from the LIDAR, the accelerometer and the RTS- smoother that uses both the LIDAR and the accelerometer. . . . . . 28 3.1 The age, height, weight and gender distributions of the participants in the study. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 3.2 Histograms of how often the participants use the different types of vehicles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 3.3 A box chart showing the distributions of the participants average speed in different segments where they were instructed to keep a con- stant speed. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 xiii List of Figures 3.4 A box chart showing the distributions of the participants standard de- viation of the speed in different segments where they were instructed to keep a constant speed. . . . . . . . . . . . . . . . . . . . . . . . . . 32 3.5 Average speed across participants during gentle braking, harsh brak- ing and unexpected braking with the four different vehicles. . . . . . 33 3.6 The braking trajectories for the harsh brake with the four different vehicles. The origin is placed at the position of the LIDAR. . . . . . 34 3.7 The distributions for the mean braking distance and the mean longi- tudinal acceleration for the braking segments. . . . . . . . . . . . . . 35 3.8 Box charts showing the distributions of the participants mean longi- tudinal acceleration on the unexpected brake with the e-scooter and the bike based on their experience. . . . . . . . . . . . . . . . . . . . 36 3.9 Box chart showing the distributions of the participants reaction time in the unexpected task. . . . . . . . . . . . . . . . . . . . . . . . . . . 36 3.10 Average speed during gentle acceleration and harsh acceleration with the four different vehicles. . . . . . . . . . . . . . . . . . . . . . . . . 37 3.11 A box chart showing the distributions of the participants mean lon- gitudinal acceleration for the gentle, and harsh acceleration. . . . . . 37 3.12 A box chart showing the distributions of the participants mean abso- lute lateral acceleration during the slalom segment. . . . . . . . . . . 38 3.13 Distributions of the mean absolute steering angle. . . . . . . . . . . . 39 3.14 Distributions of the mean absolute steering rate. . . . . . . . . . . . . 39 3.15 Distributions of the mean absolute roll angle. . . . . . . . . . . . . . 40 3.16 Distributions of the mean absolute roll rate. . . . . . . . . . . . . . . 40 3.17 Distributions of the correlations between roll rate and steering rate in terms of linear fit and time delay. . . . . . . . . . . . . . . . . . . . 41 3.18 Average absolute pitch angle and pitch rate at high speed, low speed and slalom with the Segway. . . . . . . . . . . . . . . . . . . . . . . . 41 3.19 Distribution of the mean absolute stick inclination rate with the Seg- way during slalom. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 3.20 A spider plot of the result of the questions about how the vehicles performed when riding. The scale is converted from very poor, poor, fair, good, very good, excellent and exceptional to 1-7 where 1 corre- sponds to very poor and 7 to exceptional. . . . . . . . . . . . . . . . . 43 3.21 A spider plot of the overall questions of the vehicles performance. The scale is converted from very poor, poor, fair, good, very good, excellent and exceptional to 1-7 where 1 corresponds to very poor and 7 to exceptional. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 xiv List of Tables 2.1 Technical specifications of PhidgetSpatial 3/3/3 1044_1B. . . . . . . 10 2.2 Performance indicators for bike, e-bike and e-scooter. Const, Acc and Dec are abbreviations for constant, acceleration and deceleration. S,M and C are abbreviations for stability, maneuverability and comfort. 19 2.3 Performance indicators for the Segway. Const, Acc and Dec are ab- breviations for constant, acceleration and deceleration. S,M and C are abbreviations for stability, maneuverability and comfort. . . . . . 20 3.1 Number of participants used for the different PIs. . . . . . . . . . . . 31 A.1 Data sets availability. Y indicating that the data is available and NX means that the data is missing with problem ID X. N1 means that the participant didn’t ride the vehicle, N2 means that LIDAR data is missing for that vehicle and N3 means that the vehicle data is missing for that vehicle. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . I xv List of Tables xvi 1 Introduction 1.1 Background In recent years, electric micro-mobility vehicles have increased in popularity dras- tically [1, 2]. Micro-mobility vehicles refers to small, lightweight vehicles which usually operates at a maximum of 25 km/h. The benefits of these vehicles are their size, maneuverability, affordability and the existence of multiple ride-sharing services. There are several different types of electric micro-mobility vehicles, for example electric bike (e-bike), electric scooter (e-scooter), electric skateboard and Segway. The e-bike is one of the most common electric micro-mobility vehicles and it still increases in popularity. According to Lee et al., bike makers and sellers are seeing increases in sales of e-bikes [3]. They note that in Germany, e-bike sales in 2018 was up by 36 percent compared to the previous year, which corresponds to nearly one million e-bikes and in the first half of 2019 another million e-bikes were sold. A different vehicle that has become very popular the last years is the electric scooter. The increase in popularity of e-scooters has also led to more injuries. According to Farley et al., the estimated visits due to e-scooter accidents in emergency depart- ments in the US has increased from 4881 visits in 2014 to 29 628 visits in 2019 [4]. A major factor of the increase in popularity of e-scooters is due to ride-sharing services that distributes the scooters in larger cities. This could be seen by reported accidents in Stockholm where e-scooter companies started to rent out e-scooters in 2018 [5]. In 2018, only four accidents related to e-scooter were reported, while in 2019 (until 19 September), 150 accidents were reported [6]. An electric micro-mobility vehicle that differs a lot from e-scooter and e-bike is the Segway. The Segway was invented in 1999 and was mostly used for tourist tours and law enforcement [7]. However, its price was comparably high, which probably is why the total sales of the original model, the Segway PT, were only 140 000 [8]. The movement of the Segway is controlled only by leaning which differs a lot from for example e-bike and e-scooter [9]. In Sweden, to be classified as bikes, electric micro-mobility vehicles must have a maximum speed of 20 km/h [10]. E-bikes are however an exception and are allowed to assist with motor power up to 25 km/h. The characteristics of electric micro- 1 1. Introduction mobility vehicles are however often very different compared to a regular bike. A regular bike is for example self-stable at some speeds while this is never the case for the e-scooter [1]. Previous participant studies have shown that the e-scooter may be behind conventional bikes in braking capabilities [2, 11]. There is, however, no conclusive evidence from a larger amount of participants. 1.2 Previous Research This thesis is a continuation of Alessio Violin’s thesis from 2020 [11]. Violin de- veloped a data collection and data analysis procedure for comparing the safety of e-bikes, e-scooters and Segways in terms of stability, comfort and maneuverability. The data collection and analysis Violin did, were based on test data from eight dif- ferent persons. The conclusion drawn from his thesis was that the Segway was easy to maneuver at low speed but increased in difficulty as speed increased based on rider experience. The e-scooter was perceived by the participants almost as safe as the e-bike with a high maneuverability level and high comfort. Poor braking capa- bilities were the factor that made the e-scooter less safe than the e-bike. The e-bike was the most stable and comfortable in general but struggled a bit with lateral mo- tion and low speed maneuvering. A more extensive testing needs to be performed to validate the results. The goal of the study done by Garman et al. was to investigate the influence of the rider kinematics and vehicle dynamics on e-scooter stability [2]. They designed a test course to simulate an urban environment which required multiple maneuvers. The maneuvers were turning, slalom, acceleration, deceleration, stopping and unex- pected braking. A commercially available e-scooter instrumented with sensors was used to measure acceleration, velocity, steering angle, roll angle and GPS location. Out of the eight participants in the study, seven gave data that could be used for the analysis. The results show that when riding straight on a flat path the e-scooter show high levels of stability. During the low speed turning and slalom maneuver it was noted that the participants had to use counterbalance with their body to stabilize the vehicle. Results needs to be validated on a larger set of participants. A similar field experiment as the one performed during this thesis has been done be- fore by Kovácsová et al. [12]. They investigated cycling performance of middle-aged versus older participants during tasks where stabilization skills were important. In their study, they specifically looked at and compared conventional bikes and e-bikes. The authors came to the conclusion that cyclists rated themselves to be better than the average cyclists of the same age. Another conclusion that was drawn was that the participants’ self-reported cycling skills were different to the actual cycling per- formance. Participants had a lower roll rate while cycling at low speed with the e-bike than with the conventional bike which might indicate difficulties with stabi- lizing the vehicle. When participants were to ride at their own pace, they adopted a higher speed on the e-bike than the conventional bike. The participants reached the desired speed faster on the e-bike than on the conventional bike while accelerat- 2 1. Introduction ing. This indicates that they need to exert more force on the conventional bike to accelerate. This indicates a lower comfort for the regular bike than on the e-bike. Miller et al. conducted an experiment to investigate the rider behavior of 20 Segway operators: ten experienced operators and ten novices[13]. The experiment specif- ically investigated the approach speed and clearance that Segway devices exhibit on encountering a variety of obstacles on the sidewalk. They concluded that both novice and experienced Segway riders were capable of traveling past the various ob- stacles. Research containing more parameters regarding the safety of the Segway is required in order to draw any conclusions regarding the general safety of the vehicle. 1.3 Research Question The research question that was investigated in this thesis is: “How safe, in terms of stability, maneuverability and rider comfort, are e-bikes, e-scooters and Segways based on sensor data and user experi- ence?” 1.4 Purpose The purpose of this thesis is to collect evidence for or against the safety level of e-bikes, e-scooters and Segways in terms of stability, maneuverability and rider com- fort. 1.5 Scope There are plenty of electric personal mobility vehicles available, for example e-bikes, e-scooters, Segways, electric skateboards, electric motorcycles and electric trikes just to name a few. In this thesis only the first three, the e-bike, e-scooter and Segway, are used in the evaluation of safety levels. The participants included in the experiments conducted in this thesis are recruited in the Gothenburg area, Sweden, and have to meet some inclusion criteria. To ensure the safety of the participants the experiments are performed in an area with little to none traffic and no hazardous tests are performed. 3 1. Introduction 4 2 Methodology 2.1 Vehicle Selection Vehicle selection was made with a large influence from last year’s thesis by Alessio Violin [11] since the vehicles used in his study were available and mounted with sen- sors. The vehicles were chosen based on popularity, possibility of mounting sensors and the ease of learning to ride. 2.1.1 E-bike The chosen e-bike is a Monark Karin 3-VXL, shown in Figure 2.1, which is powered by an EGOING-motor in the front wheel. The EGOING-motor helps the rider reach a maximum speed of 25 km/h. This e-bike was chosen because of its ease of use and it is an common model. Monark Karin 3-VXl has 5 different levels (1-5) of electric assistance from the motor. The higher the level is, the higher the electric assistance is. The e-bike is equipped with a coaster brake for the back wheel and a disc brake for the front wheel. Figure 2.1: Monark Karin 3-VXL electric bike. 5 2. Methodology As a reference, the e-bike was also used in the experiment as a conventional bike by turning off the electric assist by setting the assistance mode of the e-bike to 0. 2.1.2 E-scooter The e-scooter that was used is the Ninebot KickScooter ES2, shown in Figure 2.2. It has a maximum speed of 25 km/h. Although this is higher than 20 km/h which is the highest speed to be classified as a bike in Sweden, it is still a commonly used model. It has two levers on the handlebar, one for acceleration and one for brak- ing. There are three different power modes available on the Ninebot Kickscooter ES2. The three modes are sport mode, standard mode and speed limit mode. The sport mode gives the most power and speed but results in lower range while the speed limit mode gives the least power and speed but results in the longest range. The standard mode is in between the sport and speed limit mode. The e-scooter is equipped with an electric brake for the front wheel and a mechanical brake for the back wheel. The mechanical brake is integrated in the rear fender where the rider steps on it to initiate the brake. Figure 2.2: Ninebot Kickscooter ES2. 6 2. Methodology 2.1.3 Segway The last electric micro-mobility vehicle that was examined was the Segway Ninebot S, shown in Figure 2.3. According to Segway’s store, it is a smart self-balancing electric transporter which is extremely portable, easy to learn and exciting to ride [14]. It is not so common on the street, but is chosen due to its interesting balancing and steering mechanism. The Segway Ninebot S has three different modes: sports mode, new rider mode and safe mode with speed limits of 19, 10 and 7 km/h re- spectively. By gently leaning forward and backward one controls the velocity of the Segway Ninebot S. This model of the Segway does not have a handlebar. Instead, the steering bar is located between the legs and to turn, one leans gently left or right against the steering bar. Figure 2.3: Segway Ninebot S. 2.2 Data Acquisition System The vehicles were equipped with sensors to collect kinematics data. The sensors mounted on the vehicles are Inertial Measurement Units (IMUs) and potentiome- ters. The potentiometer was used to measure the steering angle and steering rate. IMUs were used to be able to measure angular velocities and accelerations of the vehicles. It was also used to estimate the orientation of the vehicles. The mount- ing locations of the sensors can be seen in Figure 2.4. All three vehicles are also equipped with a data logger which is used to record the sensor data. 7 2. Methodology (a) E-bike/bike (b) E-scooter (c) Segway Figure 2.4: Vehicle instrumentation. 8 2. Methodology A light detection and ranging sensor (LIDAR) was used to track the planar motion of the vehicles. During the experiments, the LIDAR was static and mounted on a tripod, 80 cm above the ground, to the side of the test area. In order to get the best view with the LIDAR it was placed in the middle of the 100-meter test track. 2.2.1 Data Logger The data logger is the device logging the data coming from the different sensors. The data logger is a Raspberry Pi 3 model B, which is a single-board computer with a Quad Core 1.2GHz Broadcom BCM2837 64bit CPU and 1GB RAM, using open-source software1 to save all the data to a USB memory stick. The open source software is written in Python and uses the robotic operating system (ROS). The data logger has two buttons, one for starting and stopping the recording of data and the other one has two functions. While the data logger is recording data, the second button works as a flag button meaning it gives an output of 1 when pressed and 0 otherwise. The flag feature was used to synchronize the data signals with the LIDAR’s data signals. The other function of the second button is to power off the data logger completely which can only be done if the data logger is not recording any data. The data logger can be seen in Figure 2.5. Figure 2.5: Data logger which is mounted on the e-bike. On top of the data logger is an IMU mounted. 2.2.2 Inertial Measurement Unit The IMU used in this thesis was the PhidgetSpatial 3/3/3 1044_1B. An IMU is used to measure acceleration, angular rates and magnetic field in three dimensions. Technical specifications for the IMU can be seen in Table 2.1. 1https://github.com/ruvigroup/div_datalogger 9 https://github.com/ruvigroup/div_datalogger 2. Methodology Accelerometer Acceleration measurement maximum range ±2.5g Acceleration measurement resolution 76µg Gyroscope Gyroscope speed maximum range ±100◦/s Gyroscope resolution 0.0031◦/s Magnetometer Magnetic field maximum range ±49.2G Magnetometer resolution 1.5mG Magnetometer noise 16mG Table 2.1: Technical specifications of PhidgetSpatial 3/3/3 1044_1B. 2.2.3 Steering Angle Sensor The e-scooter and the e-bike are both instrumented with a steering angle sensor in the form of a potentiometer which can be seen in Figure 2.6. A belt system connected to a wheel mounted on the potentiometer and the handlebar makes the internal resistance of the potentiometer vary depending on the angle of the handle- bar. An analog to digital converter (ADC) was used to measure the voltage across the poles of the potentiometer, induced by the varying internal resistance. The volt- age is then converted into steering angle by linear fitting. The ADC used was a 10 bit ADC connected to the Raspberry Pi through a serial peripheral interface (SPI) connection. (a) E-bike (b) E-scooter Figure 2.6: Steering angle sensor mounted on the steering stem of the e-bike and e-scooter. 10 2. Methodology To convert the voltage from the potentiometer to a steering angle, a calibration recording was made. The voltage was recorded when the handlebar was moved from -30 degrees to 30 degrees in 5-degree steps where 0 degrees are when the handlebar is straight. The linear function from the voltage to the corresponding angles that had the least squared error was then used to convert voltage to degrees. This calibration was made both for the e-scooter and e-bike. Due to the lack of a handlebar on the Segway, no potentiometer was mounted. On the Segway, the steering angle would compare to the stick inclination angle which was measured using the IMU mounted on the top of the steering stick. The IMU encased in a plastic case can be seen in Figure 2.7. Figure 2.7: IMU mounted on the top of the steering stick of the Segway. 2.2.4 LIDAR In order to record the motion of the vehicles, a LIDAR sensor was used. The LIDAR used was the Hokuyo UXM-30LXH-EWA which has a guaranteed detection range of 30 m and a maximum detection range of 120 m. The LIDAR has a scanning angle of 190° and an angular resolution of 0.125°. The logging platform for the LIDAR was run on a Raspberry Pi 3 Model B implemented as a ROS package. The logging software is the same as for the data loggers on the vehicles. A button was connected to the Raspberry Pi which was used to synchronize the data from the LIDAR and the vehicles. The button was also used to indicate when a maneuver started and ended. To start the LIDAR, a web interface is used where you can either visualize the LIDAR data in real time or record the data. The LIDAR mounted on a tripod 11 2. Methodology can be seen in Figure 2.8. Figure 2.8: LIDAR mounted on a tripod. 2.3 Participant Selection Participants that met the requirements in the list below were invited to the study. • Be at least 160 cm tall. • A maximum weight of 85 kg. • Be 18-50 years old. • Be able to ride a bicycle. • Do not have any physical disabilities. • Have not suffered from serious traffic accidents. • No symptoms of Covid-19 in the last two weeks before the study. The height criteria and the weight criteria were from limitations of the vehicles. The bicycle used was not suited for participants below 160 cm and the Segway had a maximum weight limit of 85 kg. The requirement, to be able to ride a bicycle, was set to have a good reference for the electric micro-mobility vehicles and to reduce the training time needed. The two requirements, not have any physical disabilities and not have suffered from serious traffic accidents were set primarily due to ethical concerns. The age limit was set to reduce the risks of injuries and to limit the number 12 2. Methodology of between-subject factors. Apart from these requirements, a gender balance was desired since both males and females use electric micro-mobility vehicles and could potentially have different riding behaviors. 2.4 Experimental Protocol All of the experiments with the participants were done on a pier located at Pump- gatan 2 in Gothenburg, Sweden. The pier is a paved road for cyclists and pedestrians with little to no traffic. This section will cover a description of the four tasks the participants performed, how the test track was set up prior to participants arriving at the scene as well as how the experiments were done in detail with each participant. 2.4.1 Description of Tasks The participants performed four different tasks on each vehicle. The different ma- neuvers that the participants performed are adapted from work by Kovácsová et al. [12] and Rasch et al. [15]. Figure 2.9: The first task which is to study the behavior of constant high speed, comfortable braking and comfortable acceleration. The first task could be seen in Figure 2.9. It starts with the rider accelerating gently until a speed between 17 km/h to 20 km/h has been reached and then continue in the same speed before braking to stop before a line in a gentle way. Then acceler- ating gently up to the interval 17-20 km/h again. This task was chosen to indicate how the different vehicles behave in high speed, an expected braking situation and accelerating. This is a common real life scenario when approaching a stop sign. A small speed interval is set to give a fair comparison of the different vehicles, but also make it easy for the rider to complete the task. The risk of setting an absolute speed is that the rider gives a lot of attention to the speed monitor to not deviate from the requested speed. 13 2. Methodology Figure 2.10: The second task which studies the behavior of constant high speed, harsh braking and harsh acceleration. The second task, which is the harsh maneuver, can be seen in Figure 2.10. It starts with the rider accelerating harshly to 17-20 km/h and maintaining that speed. The rider should then brake harshly and come to a complete stop before a line. Then they should proceed with accelerating harshly up to 17-20 km/h again and maintain that speed until they reach the finish line. In a real life scenario this could represent a person that perceives a stop signal too late due to not being attentive. This task also shows the braking and accelerating capabilities of the different vehicles. Figure 2.11: The third task which studied the behavior of a harsh acceleration and an unexpected harsh brake. The third task could be seen in Figure 2.11. It starts with accelerating to the interval 17-20 km/h. The rider then keeps a constant speed until it hears a sound signal, which is that one of the experimenters shouts stop. When the rider hears the sound signal they should brake harshly. This task represents an unexpected stop. In a real life scenario it is important to have a short braking distance to avoid a potential crash, but also that the braking is stable to avoid falling. 14 2. Methodology Figure 2.12: The fourth task which studies the behavior during consecutive turns. The fourth and final task is shown in Figure 2.12. Here the rider should accelerate to 7-10 km/h and try to maintain this speed while completing a slalom course around four cones. Here the performance of the vehicles in turns are studied. This task could represent passing obstacles or pedestrians in a real setting. 2.4.2 Set Up Three lines were drawn on the ground using chalk to indicate start, brake and finish locations. The start line is where all the tasks were started from and the finish line is the end of the test track. The brake line is the line used in the gentle and harsh maneuver to indicate where the participants should have come to a complete stop. Four small cones were placed 3 meters apart, one by the brake line, two before the brake line and one after the brake line. The LIDAR was positioned parallel to the traffic cones aligned with the brake line to best capture the motion of the vehicles in the most important part of the maneuvers which is acceleration, deceleration and slalom. An illustration of the test area can be seen in Figure 2.13 and a photo of the middle part of the test area can be seen in Figure 2.14. Figure 2.13: Illustration of the test area. 15 2. Methodology Figure 2.14: Photo of the middle part of the test area. Before each participant arrived, the LIDAR was tested to see that it was aligned correctly and worked properly. Lamp posts on the pier was used to make sure the LIDAR was parallel to the road using the visualize function of the web application. To reduce the spread of COVID-19 the handlebars of the vehicles and the seat of the bike were disinfected and the experimenters had visors or protective masks. 2.4.3 Main Experiment When a participant arrived at the test location, a short introduction was given about the experiment. The participant was then asked to read through and sign a consent form. After the consent form was signed, the experimenter and the participant went through the experimental protocol together. When all the required information were given to the participant, they were equipped with a helmet and optional protectors for their knees, elbows and hands. The par- ticipants were then asked to try out all four vehicles (bike, e-bike, e-scooter and Segway). If any of the participants did not feel safe or felt uncomfortable riding any of the vehicles, that vehicle was skipped in the experiment. Same goes for the four different tasks. If any of the tasks felt uncomfortable for the participant, the task was skipped. After the participant had tried out all the vehicles, the main experiment could start. The order in which each of the participants were to ride each vehicle and task was randomized. This was done to reduce any potential learning effect. Before starting the tasks with each vehicle, there were some final set-up steps needed. The recording of the LIDAR and the data logger on the vehicle were started. When the recording of the on-board sensors on the vehicle, through the data logger, was started, the 16 2. Methodology vehicle was kept in a stationary position for a couple of seconds. This was to give the data logger enough time to calibrate the gyroscope. After the calibration was done the flag button was pressed simultaneously on the data logger of the vehicle and on the LIDAR to be able to sync the different sensors. Before each task, the participant was given oral instructions on how to perform the task which is shown in appendix C. The experimenter operating the LIDAR did a short press on the flag button when the participant started each task and a long press whenever the participant finished the task. For the unexpected task, a but- ton press was also done whenever the signal was given to brake. For the e-bike, a set electric assistance and mechanical gear was used for the different tasks. For the slalom task the electric assistance was set to 2 and the mechanical gear was set to 1. In the other three tasks, the electric assistance was set to 4 and the me- chanical gear to 2. The e-scooter and Segway always had the highest power mode on. After the participant had completed all the tasks with a vehicle, the data logger on the vehicle and the LIDAR was stopped. Then the process restarted with starting the data logger on the next vehicle and so on. 2.4.4 Questionnaire When the participants were done with the riding tasks, they were asked to fill in a questionnaire about their experience during the riding tasks as well as when they tried out all the vehicles. The questionnaire can be seen in appendix B. It starts with questions regarding the demographics and prior experiences of the participant. It continues with questions about how they felt the different vehicles performed in 11 different scenarios. The questionnaire ends with four questions regarding the overall comfort, maneuverability, stability and safety. All of the questions regarding the performance of the vehicles and the overall questions are questions with a 7-grade scale. The scale can be seen below: 1. Very Poor 2. Poor 3. Fair 4. Good 5. Very Good 6. Excellent 7. Exceptional After every question there is an optional field for adding comments. If the participant chose not to ride any of the vehicles, the answers related to the skipped vehicles were removed. 2.5 Performance Indicators Performance indicators (PIs) are information collected at regular intervals to track the performance of a system. In this thesis, a list of different PIs are being measured 17 2. Methodology and calculated in order to assess the safety of the e-vehicles in the dimensions sta- bility, maneuverability and rider comfort. A high stability indicates that the vehicle does not need a lot of correction by the user when riding. A high maneuverability indicates that the vehicle is easy to accelerate, decelerate and turn with. A high comfort indicates that the rider feels comfortable, does not need a lot of physical effort to maneuver the vehicle and is not subject to excessive force. The PIs are pa- rameters that can relate vehicle kinematics to one or more safety dimensions. Each task is split up into smaller segments. For the gentle and harsh maneuvers, there are three segments: 1) the constant seg- ment, 2) the acceleration segment and 3) the deceleration segment. The constant segment is the part of the task where the rider is trying to maintain a constant speed of 17-20 km/h. The constant segment will occur twice, both before the acceleration segment and after the deceleration segment. The slalom task is split up into two segments. The first segment is the constant seg- ment where the rider tries to maintain a constant speed of 7-10 km/h which occurs both before and after the slalom segment. The slalom segment is the part of the task where the rider is riding slalom around the traffic cones. In order to examine the reactions and braking capabilities during the unexpected task, it is split up into three segments. First, there is the constant segment where the rider tries to maintain a constant speed of 17-20 km/h. The second segment is the reaction segment which is from when the signal is given until the rider starts to brake. The third, and final segment, is the deceleration segment which is just like in the gentle and harsh maneuver the segment where the rider decelerates. The full list of PIs for the bike, e-bike and e-scooter can be seen in Table 2.2. The full list of PIs for the Segway can be seen in Table 2.3. 18 2. Methodology Signal Performance Indicator Segment Interpretation Steering angle (deg) Mean absolute steering angle Const, Slalom S,M Steering rate (deg/s) Mean absolute steering rate Const, Slalom S Roll angle (deg) Mean absolute roll angle Const, Slalom S Roll rate (deg/s) Mean absolute roll rate Const, Slalom S Speed (km/h) Mean speed Const, Slalom M,C Speed (km/h) Standard deviation Const, Slalom M,C Time (s) Time Acc, Dec, Reaction M Longitudinal acceleration (m/s2) Mean longitudinal acceleration Acc, Dec M,C Lateral acceleration (m/s2) Mean absolute lateral acceleration Slalom C Distance (m) Braking distance Dec M Steering rate (deg/s), Roll rate (deg/s) R2 of linear fit Slalom S Steering rate (deg/s), Roll rate (deg/s) Time delay between roll rate and steering rate Slalom M Table 2.2: Performance indicators for bike, e-bike and e-scooter. Const, Acc and Dec are abbreviations for constant, acceleration and deceleration. S,M and C are abbreviations for stability, maneuverability and comfort. 19 2. Methodology Signal Performance Indicator Segment Interpretation Stick inclination rate (deg/s) Mean absolute stick inclination rate Slalom S,M Pitch angle (deg) Mean absolute pitch angle Const, Slalom C Pitch rate (deg/s) Mean absolute pitch rate Const, Slalom S Speed (km/h) Mean speed Const, Slalom M,C Speed (km/h) Standard deviation Const, Slalom S,M,C Time (s) Time Acc, Dec, Reaction M Longitudinal acceleration (m/s2) Mean longitudinal acceleration Acc, Dec M,C Lateral acceleration (m/s2) Mean absolute lateral acceleration Slalom C Distance (m) Braking distance Dec M Table 2.3: Performance indicators for the Segway. Const, Acc and Dec are abbre- viations for constant, acceleration and deceleration. S,M and C are abbreviations for stability, maneuverability and comfort. 20 2. Methodology 2.5.1 Speed The mean speed and the standard deviation of the speed are used as PIs for the constant and slalom segment. A too low mean speed may indicate that the rider is not comfortable enough to ride at a higher speed while a too high speed may indicate that the rider has trouble with maneuvering the vehicle. The standard deviation of the speed is a measurement of both maneuverability and comfort. If the standard deviation of the speed is low it indicates that it is easy to keep a constant speed which indicates a high maneuverability. It also indicates a high comfort since it indicates that the user need to correct the speed less. For the Segway, the standard deviation is also related to stability since trouble with balance will lead to changes in speed. 2.5.2 Braking The braking distance, the braking time and the mean longitudinal acceleration are used as PIs for the deceleration segment. A short braking distance or time and a high deceleration for the unexpected and harsh brake indicates that it is possible to brake fast which indicates a high maneuverability. A high deceleration may however also indicate less comfort. The reaction time in the unexpected task is chosen as a PI to indicate maneuver- ability since a high reaction time indicates that the rider needs more time to initiate the braking. 2.5.3 Acceleration The acceleration time and the mean longitudinal acceleration are used as PIs for the acceleration segment. A short acceleration time and a high acceleration for the harsh acceleration indicates that it is possible to accelerate fast which indicates a high maneuverability. A high acceleration may however also indicate less comfort. For the slalom segment the mean absolute lateral acceleration is used as a PI since a high lateral acceleration may indicate less comfort. 2.5.4 Steering and Leaning 2.5.4.1 E-bike, Bike and E-scooter The mean absolute steering angle, steering rate, roll angle and roll rate are used as PIs for the constant and slalom segment. The peak cross-correlation between steering rate and roll rate and the time delay between steering rate and roll rate are used as PIs for the slalom segment. A high mean absolute steering angle, steering rate, roll angle and roll rate indicates that the rider needs to do a lot of correction to maneuver the vehicle, hence a low stability. For the slalom segment a low steering angle also indicates high maneuverability since less steering is needed to pass the cones. The peak cross-correlation between the steering rate and the roll rate shows 21 2. Methodology how correlated the two control input from the rider are. A high correlation indicates that the user have control of the maneuver and therefore high stability. A low time delay between the roll rate and the steering rate indicates that the vehicle is responsive, hence high maneuverability. 2.5.4.2 Segway The pitch angle, pitch rate and the stick inclination rate are used as PIs for the constant and slalom segment. The mean absolute pitch angle is related to comfort since it may be less comfortable to lean forward than to stand in a upright position. A high pitch rate may indicate that the rider does not have the stability to stand in a stable way on the Segway. The same reasoning can be done for the standard deviation of the speed. If the speed changes a lot it indicates that the posture also changes a lot which may be a sign that the Segway is not stable. A low stick inclination rate during the slalom may indicate that the rider does not need to do fast correction to go through the slalom course and is therefore a sign of stability. 2.5.5 Box Plot The PIs will mostly be presented in box charts that represent the distribution of the data. A description of the box chart is shown in Figure 2.15 where the line inside of the box is the sample median. The top and bottom edges of the box are the upper and lower quartiles, respectively. Interquartile range (IQR) is the distance between the upper and lower quartiles. The upper quartile corresponds to the 0.75 quantile and the lower quartile corresponds to the 0.25 quantile. To be classified as an outlier the value has to be more than 1.5 · IQR away from the top or bottom of the box. Values that are outside the quartiles but not an outlier are represented by the whiskers which in turn shows the nonoutlier maximum and nonoutlier minimum. The notch is used to see the significance of difference of medians. If notches of two boxes do not overlap, it means that they have different medians at the 5% significance level. The significance level is based on a normal distribution assumption, but can also be used to compare other distributions. If m is the median and n is the number of data points, the notch regions top and bottom edges correspond to m+ (1.57 · IQR)/ √ n and m− (1.57 · IQR)/ √ n, respectively. 22 2. Methodology Figure 2.15: Description of a box chart. 2.6 Data Analysis 2.6.1 Calibration To be able to use the data from the IMUs to compute the PIs, the orientation with respect to the vehicle needs to be known or estimated. The direction of the gravity in the IMU frame is computed by collecting readings from the accelerometer, while the vehicle is set in an upright position with the use of a level. From this a rotation matrix that aligns the z-axis of the IMU with the vehicles vertical axis could be computed by solving Equation (2.1), where g is the gravitational constant, R1 is the rotation matrix and, ax, ay and az are the readings from the accelerometer.0 0 g  = R1 axay az  (2.1) The next step is to find the rotation matrix that aligns the x-axis of the IMU with the lateral direction of the vehicle and the y-axis with the longitudinal direction. This was done by leaning the vehicles such that the direction of the gravity is directed only in the lateral and the vertical direction of the vehicle. The second rotation matrix could then be computed from the rotated accelerometer reading by solving Equation (2.2). In this equation k is an arbitrarily constant, R1 is the rotation matrix from Equation (2.1), R2 is the second rotation matrix and, ax and ay are readings from the accelerometer. By multiplying these two rotation matrices a rota- tion matrix from the IMU frame to the vehicle frame was estimated. This rotation 23 2. Methodology matrix was then applied to the acceleration readings from the accelerometer and the angular velocity readings from the gyroscope. 0 k 0  = R2R1 axay 0  (2.2) 2.6.2 Post Processing In order to be able to compute the PIs for the vehicles, the data collected during the experiments needed to be processed. This section will cover the filtering methods used, how the tasks were split up into the segments mentioned in Section 2.5, how the orientation of the vehicles were estimated, and how the LIDAR data, together with the IMU data, were used to estimate position and speed. The process of the post processing can be seen in Figure 2.16. Figure 2.16: Process description for post processing. 2.6.2.1 Syncing the Times The data logger on the vehicles and the logging device on the LIDAR were instru- mented with a real-time clock. The real-time clocks however suffered from drift over time. All the logging devices had a connected flag button. Before starting the tests with a vehicle, the button connected to the vehicle was pressed at the same time as the button connected to the LIDAR. For some participants, the syncing was not correct due to human errors and hard- ware problems. The human errors were missing to press a button or not pressing the buttons at the same time. Hardware problems was that the button press was not registered due to loose wire connection. To sync the signals, the longitudinal acceleration of the harsh task was derived from the LIDAR measurements. The time lag between the LIDAR measurements and the IMU measurements was then estimated with the finddelay function in MATLAB with the acceleration from the LIDAR and the accelerometer of the IMU. The function finds the delay that gives the highest cross correlation between the two signals. 2.6.2.2 Filtering The steering angle, steering rate, angular velocity and orientation angles were all filtered with a low-pass filter. The low-pass filter used was a zero-phase Butterworth 24 2. Methodology filter [16] with a passband of 0-1 Hz and a stopband of 1-1.5 Hz. 2.6.2.3 Segmentation The start and end of each maneuver were indicated by a flag button and could therefore easily be extracted. Each maneuver also consists of different segments. These are not indicated by any button press and instead based on the signal data. The first segment of all the maneuvers is the constant segment which started after the first 20 traveled meters, so that the rider had enough time to accelerate to the given speed. For the braking tasks, the next segment is the braking segment. It starts when going under 16 km/h the last time before the stop for the bike, e-bike and the e-scooter and ends when the speed is under 2.5 km/h. For the Segway, it starts when the rider is going under 12 km/h the last time before the stop since the participant had a lower speed with the Segway. The unexpected task ends after the stop, but it also has a reaction segment which starts when the signal is shouted. After the vehicle decelerates to a speed that is 1 km/h less than the speed at the time of signal the reaction segment ends. For the gentle and harsh task, the third segment is the acceleration segment which starts when the speed goes above 2.5 km/h and ends at 16 km/h for e-bike, bike, and e-scooter, and 12 km/h for the Segway. After that, a second constant segment starts which ends 10 meters before the finish line to avoid collecting data from braking or steering that some participants did just before the finish line. Examples of the segmentation for the braking tasks is shown in Figure 2.17. (a) Unexpected task. (b) Harsh/Gentle task. Figure 2.17: Example of the segmentation for the braking tasks. The slalom task is segmented solely based on position. After the first constant seg- ment there is a slalom segment. The slalom segment starts 4 meters in front of the first cone and ends 4 meters after the last cone. The second constant segment starts directly after the slalom segment and ends 10 meter before the finish line. An 25 2. Methodology example of the segmentation for the slalom task is shown in Figure 2.18. Figure 2.18: An example of the segmentation for the slalom task. The origin is the position of the LIDAR. 2.6.2.4 Steering Angle Correction A mean correction of the steering angle was made for each maneuver to correct for potential drift. During the constant segment it was assumed that the mean steer angle was zero. The mean steer angle of the constant segment was therefore removed from the steer angle measurements for each task. 2.6.2.5 Orientation Estimation The orientation of the vehicles was estimated by using a Madgwick filter [17] with the measurements from the gyroscope and accelerometer. The magnetometer values were not used since the values were heavily affected by disturbances, even after calibration. This is probably due to soft iron disturbances from the electronics on the vehicle which is very hard to compensate for. The magnetometer is also mostly used for the yaw angle estimate which could be estimated from the LIDAR measurements instead. 2.6.2.6 LIDAR Data Processing The LIDAR data consist of a set of 2D points. To extract the position of the vehicle, the LIDAR data were clustered. The detected view was limited to the test area to avoid potential clusters from the environment. The limits were set to ±50 meters in the longitudinal direction, and between two and nine meters in the lateral direction, where the position of the LIDAR is placed at the origin. The clustering method that was used was DBSCAN which is a clustering algorithm that clusters 26 2. Methodology data based on the density and can discover clusters of arbitrary shape [18]. The center of the cluster representing the vehicle was then interpreted as the center of the vehicle. A first estimate of the speed of the vehicle was then estimated from the LIDAR data by measuring the change of position of the cluster center over time. A better estimation of the speed was then computed by fusing the estimation from the LIDAR with the accelerometer data. 2.6.2.7 Rauch-Tung-Striebel Smoother The Rauch-Tung-Striebel (RTS) smoother [19] was used to estimate the position and the speed of the rider. For the gentle, harsh and unexpected task, there should be no turning and therefore a one-dimensional model was used. The prediction model is shown in (2.3) where the distance (d) and speed (v) represent the state vector and the longitudinal acceleration (a) from the accelerometer on the vehicle is the input. The sampling period is denoted as T . The measurement model is shown in (2.4) and uses the distance (DL) and speed (VL) acquired from the LIDAR measurements. [ dk vk ] = [ 1 T 0 1 ] [ dk−1 vk−1 ] + [ T 2 2 T ] ak +Q1 (2.3) [ DL,k VL,k ] = [ 1 0 0 1 ] [ dk vk ] +R1 (2.4) In the slalom task, the rider turns and therefore a two-dimensional model was used. The prediction model for the slalom task is shown in (2.5) where the state vector includes the position (x, y), the linear speed (v) and the heading (φ). The position of the LIDAR is set to the origin. The position in the longitudinal direction of the track is described with x. The position in the lateral direction is described with y. The longitudinal acceleration (a) from the accelerometer and the yaw rate (ω) from the gyroscope are treated as inputs. The sampling period is denoted as T . The measurement model for the slalom model is shown in (2.6) and uses the position (XL, YL) and speed (VL) acquired from the LIDAR measurements.  xk yk vk φk  =  1 0 T cos(φk−1) −Tvk−1 sin(φk−1) 0 1 T sin(φk−1) Tvk−1 cos(φk−1) 0 0 1 0 0 0 0 1  +  0 0 0 0 T 0 0 T  [ ak ωk ] +Q2 (2.5) XL,k YL,k VL,k  = 1 0 0 0 0 1 0 0 0 0 1 0   xk yk vk φk  +R2 (2.6) Q and R are matrices that describe the noise. Q represents the uncertainty of the prediction model and R represents the insecurity of the measurements. These ma- trices were tuned to give a stable result and are shown in (2.7) and (2.8) where σ was set to 10 for the Segway and 1 for the other vehicles. The sampling period is 27 2. Methodology denoted as T . Q1 = σ [ T 4 4 T 3 2 T 3 2 T 2 ] Q2 = 10−4  10 0 0 0 0 10 0 0 0 0 100 0 0 0 100 π 180  (2.7) R1 = [ 10 0 0 1 ] R2 = 0.1 0 0 0 0.5 0 0 0 1  (2.8) An example of the result of the RTS smoother that was used for the gentle, harsh and unexpected task is shown in Figure 2.19. It shows that by just using the LIDAR measurements, the estimates are very noisy. By simply integrating the longitudinal acceleration from the accelerometer, drift occurs. By combining the LIDAR and the accelerometer, the estimate is much smoother than by just using the LIDAR and with no notable drift as in the case of just using the accelerometer. 0 5 10 15 20 25 Time (s) 0 5 10 15 20 25 S p e e d ( k m /h ) Integrated from accelerometer Differientated from Lidar pos RTS-smoothed Figure 2.19: A speed estimate from the LIDAR, the accelerometer and the RTS- smoother that uses both the LIDAR and the accelerometer. 28 3 Results A total of 34 participants took part in the study and conducted the experiments. The demographics of the participants from the experiments can be seen in Figure 3.1. The mean age of the participants were 25.3, the mean height were 175.1 cm and the mean weight were 71.5kg. There were 25 male participants and 9 female participants. Figure 3.1: The age, height, weight and gender distributions of the participants in the study. One of the questions in the questionnaire was about the participants prior experience with riding the different vehicles. Specifically the question was how often they used the different vehicles. The results from the prior experience question can be seen in Figure 3.2. For the Segway, it can be seen that all participants, except one, never 29 3. Results ride it and 26 participants never ride the e-bike. For the bike, most participants ride it a few days per week but the spread is even through all options except for one participant who never rides a bike. There were 12 participants who never ride an e-scooter and the rest of the participants were split up between the other options. When there were questionnaire entries related to the Segway and the participant did not ride the Segway during the experiments, those entries were removed. This is the reason why there are fewer participants represented in the Segway prior experience. Figure 3.2: Histograms of how often the participants use the different types of vehicles. 3.1 Data Overview Due to various reasons, some data from the conducted experiments are missing. A list of the data set availability can be seen in Table A.1 in the appendix. There were 9 participants who did not feel comfortable riding the Segway and the questionnaire questions regarding the Segway were therefore skipped for these participants. The data from the other vehicles were however still used for these participants. One of the participants had an accident while riding the first vehicle in the experiments, the regular bike. The experiment was canceled for that participant and that par- ticipant did not complete any task with the other three vehicles nor filled in the 30 3. Results questionnaire. The number of participants that had useful data for the different performance indicators is shown in Table 3.1. The data that was not useful were data where either the participant did not reach the desired speed, the participant did not come to a complete stop or there were a problem with the potentiometer belt. E-bike Bike E-scooter Segway Longitudinal acceleration gentle 26 22 26 13 harsh 26 25 28 13 Deceleration gentle 26 18 20 11 harsh 26 25 28 14 unexpected 28 29 28 13 Lateral acceleration 28 29 28 14 Speed constant 28 29 28 14 Braking distance 28 29 28 14 Reaction time 28 29 28 13 Roll/Pitch angle/rate 28 29 28 14 Steering angle/rate 25 25 28 N/A R2/Time delay steering-roll 25 25 28 N/A Stick inclination rate N/A N/A N/A 14 Table 3.1: Number of participants used for the different PIs. 3.2 Performance Indicators In this section, the results of the PIs are shown. The results for the constant segment on the gentle, harsh and unexpected tasks were very similar. The results for the constant segment are therefore limited to the constant segment on the harsh task and on the slalom task and will be referred to as high speed and low speed. 3.2.1 Speed In Figure 3.3, it could be seen that the e-bike has the highest mean speed on the high-speed constant segment. The Segway which did not have a speedometer and a speed limit at 19 km/h has a considerably lower mean speed than the other vehicles in the high speed segment. It also has a lower mean speed in the low speed segment and slalom segment. It could also be seen that the mean speed of all vehicles is lower in the slalom segment than on the low speed segment, even if the participants were instructed to keep the same speed in these segments. 31 3. Results High Speed Low Speed Slalom 0 10 20 S p e e d ( k m /h ) E-Bike Bike E-Scooter Segway Figure 3.3: A box chart showing the distributions of the participants average speed in different segments where they were instructed to keep a constant speed. High Speed Low Speed Slalom 0 1 2 3 4 5 S ta n d a rd D e v ia ti o n S p e e d ( k m /h ) E-Bike Bike E-Scooter Segway Figure 3.4: A box chart showing the distributions of the participants standard deviation of the speed in different segments where they were instructed to keep a constant speed. Figure 3.4 shows the standard deviation of the speed when the participants were instructed to keep a constant speed. The results can not show with 95 % certainty that there is any difference in the median between the vehicles. The median is however higher on the Segway than on the other vehicles, but the confidence interval is also very large. The standard deviation however decreases with speed for the e- bike, bike and e-scooter. It also seems to decrease for the Segway, but this is not verified with a 95 % confidence interval. The standard deviation for the Segway has 32 3. Results the largest spread meaning that the participants managed to keep a constant speed differently from each other. 3.2.2 Braking -5 -4 -3 -2 -1 0 Time (s) 5 10 15 20 S p e e d ( k m /h ) E-Bike Bike E-Scooter Segway (a) Average speed during gentle braking. -5 -4 -3 -2 -1 0 Time (s) 5 10 15 20 S p e e d ( k m /h ) E-Bike Bike E-Scooter Segway (b) Average speed during harsh braking. -5 -4 -3 -2 -1 0 Time (s) 5 10 15 20 S p e e d ( k m /h ) E-Bike Bike E-Scooter Segway (c) Average speed during unexpected braking. Figure 3.5: Average speed across participants during gentle braking, harsh braking and unexpected braking with the four different vehicles. In Figure 3.5, the average speed across participants during the gentle braking, harsh braking and the unexpected braking are shown. The Segway has the slowest braking of the four vehicles followed by the e-scooter. The bike and e-bike have similar braking, but the initial speed is higher on the e-bike, especially during gentle braking. 33 3. Results -15 -10 -5 0 5 10 Longitudinal position (m) 3 4 5 6 7 8 9 L a te ra l p o s it io n ( m ) Stop Line Start Position Brake Trajectory Stop Position (a) Braking trajectories for the bike. -15 -10 -5 0 5 10 Longitudinal position (m) 3 4 5 6 7 8 9 L a te ra l p o s it io n ( m ) Stop Line Start Position Brake Trajectory Stop Position (b) Braking trajectories for the e-bike. -15 -10 -5 0 5 10 Longitudinal position (m) 3 4 5 6 7 8 9 L a te ra l p o s it io n ( m ) Stop Line Start Position Brake Trajectory Stop Position (c) Braking trajectories for the e-scooter. -15 -10 -5 0 5 10 Longitudinal position (m) 3 4 5 6 7 8 9 L a te ra l p o s it io n ( m ) Stop Line Start Position Brake Trajectory Stop Position (d) Braking trajectories for the Segway. Figure 3.6: The braking trajectories for the harsh brake with the four different vehicles. The origin is placed at the position of the LIDAR. When performing the harsh brake, a lot of participant passed the stop line before reaching a complete stop with the e-scooter which could be seen in Figure 3.6c. This was also the case for a large proportion of the harsh brake with the Segway which is shown in Figure 3.6d. For the e-bike and bike, there were however only a few participants that passed the line before stopping which is seen in Figure 3.6a and 3.6b. 34 3. Results Gentle Harsh Unexpected 0 12 24 D is ta n c e ( m ) E-Bike Bike E-Scooter Segway (a) A box chart showing the distributions of the participants braking distance for the gentle, harsh and unexpected brake. Gentle Harsh Unexpected -7 -6 -5 -4 -3 -2 -1 0 L o n g it u d in a l A c c e le ra ti o n ( m /s 2 ) E-Bike Bike E-Scooter Segway (b) A box chart showing the distributions of the participants mean longitudinal ac- celeration for the gentle, harsh and unex- pected brake. Figure 3.7: The distributions for the mean braking distance and the mean longi- tudinal acceleration for the braking segments. In Figure 3.7a, the braking distance distributions between the different participants are shown for the gentle, harsh and unexpected brake. It could be seen that the e-bike and bike, that have the same brakes, have similar distributions for the harsh and the unexpected brake. The e-scooter has a longer braking distance than both the e-bike and the bike. The Segway starts braking from a considerably lower speed, but still has a similar braking distance as the other vehicles on the harsh and unex- pected brake. For the gentle brake, the Segway has a shorter braking distance than the e-scooter and the bike. The e-bike seems to have a shorter braking distance than the bike and the e-scooter, but this is not verified with a 95 % certainty. The e-bike and the bike have very similar deceleration values, which is shown in Fig- ure 3.7b. They are also the vehicles on which the participants braked the hardest in the gentle, harsh and unexpected brake. The braking however differ a lot between the participants, especially for the harsh and the unexpected brake. By sorting the participants by how often they use a bike, it could be seen that the participant that bikes often, brakes harder. This is shown with the box charts in Figure 3.8a. There is however no clear trend between experience and braking deceleration with the e- scooter, which is shown in Figure 3.8b. The braking of the e-scooter was slower than the e-bike and the bike, but faster than with the Segway. The harsh and the unexpected brake distributions are very similar. 35 3. Results Never Few days per y ear Few days per m onth Few days per w eek Everyday -6 -4 -2 0 L o n g it u d in a l A c c ( m /s 2 ) (a) Unexpected brake bike. Never Few days per y ear Few days per m onth Few days per w eek Everyday -6 -4 -2 0 L o n g it u d in a l A c c ( m /s 2 ) (b) Unexpected brake e-scooter. Figure 3.8: Box charts showing the distributions of the participants mean longi- tudinal acceleration on the unexpected brake with the e-scooter and the bike based on their experience. 0 0.2 0.4 0.6 0.8 1 1.2 1.4 T im e ( s ) E-Bike Bike E-Scooter Segway Figure 3.9: Box chart showing the distributions of the participants reaction time in the unexpected task. The reaction time of the participants during the unexpected brake are very similar for the e-bike, bike and the Segway which is shown in Figure 3.9. The participants however have a significantly higher median reaction time with the e-scooter than with the other vehicles. 36 3. Results 3.2.3 Accelerating 0 2 4 6 8 Time (s) 5 10 15 20 S p e e d ( k m /h ) E-Bike Bike E-Scooter Segway (a) Average speed during gentle accelera- tion. 0 2 4 6 8 Time (s) 5 10 15 20 S p e e d ( k m /h ) E-Bike Bike E-Scooter Segway (b) Average speed during harsh accelera- tion. Figure 3.10: Average speed during gentle acceleration and harsh acceleration with the four different vehicles. From Figure 3.10, it could be seen that the bike and the e-bike accelerate very similarly in the beginning, but after a while, the e-bike accelerates faster than the bike, both in the gentle acceleration and the harsh acceleration. The e-scooter and the Segway have the fastest acceleration in the beginning, but after a few seconds, the e-bike accelerates faster. This leads to that the e-bike have the highest speed after 4.3 seconds for the gentle acceleration and after 2.5 seconds for the harsh task. Gentle Harsh 0 0.5 1 1.5 2 L o n g it u d in a l A c c e le ra ti o n ( m /s 2 ) E-Bike Bike E-Scooter Segway Figure 3.11: A box chart showing the distributions of the participants mean lon- gitudinal acceleration for the gentle, and harsh acceleration. 37 3. Results In Figure 3.11, the distributions of the mean longitudinal acceleration when acceler- ating are shown. Both the Segway and the e-bike have a median of around 1 m/s2 during the harsh acceleration. The Segway however has a large interval which shows that the participants accelerated very differently with it. The e-scooter and the bike have a smaller median acceleration than the e-bike for the harsh task. For the gentle task, the bike has a smaller mean acceleration than both the e-bike and the Segway. 0 0.5 1 1.5 2 L a te ra l A c c e le ra ti o n ( m /s 2 ) E-Bike Bike E-Scooter Segway Figure 3.12: A box chart showing the distributions of the participants mean ab- solute lateral acceleration during the slalom segment. In Figure 3.12, the distributions of the mean absolute lateral acceleration during the slalom segment, are shown. From the figure it is shown that the Segway has a significantly lower lateral acceleration than the other vehicles. There is no significant difference in lateral acceleration between the e-bike, the bike and the e-scooter. 38 3. Results 3.2.4 Steering and Leaning 3.2.4.1 E-bike, Bike and E-scooter 0 5 10 15 20 S te e ri n g a n g le ( d e g ) E-Bike Bike E-Scooter (a) Mean absolute steering angle during slalom. High Speed Low Speed 0 1 2 3 4 5 S te e ri n g A n g le ( d e g ) E-Bike Bike E-Scooter (b) Mean absolute steering angle when riding straight. Figure 3.13: Distributions of the mean absolute steering angle. In Figure 3.13, the distributions of the mean absolute steering angle is shown. It could be seen that the median of the e-scooter is significantly lower than both the e-bike and the bike in the slalom segment. In the high speed segment, the e-bike and the e-scooter have a significantly lower median than the bike. In the low speed segment, there is no significant difference of the median between the vehicles even if it indicates that it is lower for the e-scooter. The mean absolute steering angles for the low speed segment is significantly larger compared to the high speed. 0 2 4 6 8 10 12 14 S te e ri n g R a te ( d e g /s ) E-Bike Bike E-Scooter (a) Mean absolute steering rate during slalom. High Speed Low Speed 0 2 4 6 8 10 12 14 S te e ri n g R a te ( d e g /s ) E-Bike Bike E-Scooter (b) Mean absolute steering rate when rid- ing straight. Figure 3.14: Distributions of the mean absolute steering rate. The average absolute steering rate distributions are shown in Figure 3.14 where the e-scooter has a significantly smaller median than the bike in all three segments. A 39 3. Results significant difference in median is also shown in the high speed segment between the e-bike and the bike where the e-bike has a lower steering rate than the bike. 0 2 4 6 8 R o ll a n g le ( d e g ) E-Bike Bike E-Scooter (a) The mean absolute roll angle during slalom. High Speed Low Speed 0 0.5 1 1.5 2 2.5 R o ll a n g le ( d e g ) E-Bike Bike E-Scooter (b) The mean absolute roll angle when riding straight. Figure 3.15: Distributions of the mean absolute roll angle. The distributions of the mean absolute roll angle are shown in Figure 3.15. There is no significant difference in the median between the vehicles, but the results indicate that the median absolute roll angle is the lowest for the e-scooter during the slalom maneuver and the lowest for the e-bike when riding straight at a low speed. There is also no significant difference in the median when riding straight at high speed and low speed. 0 5 10 15 20 R o ll R a te ( d e g /s ) E-Bike Bike E-Scooter (a) The mean absolute roll rate during slalom. High Speed Low Speed 2 4 6 8 R o ll R a te ( d e g /s ) E-Bike Bike E-Scooter (b) The mean absolute roll rate when rid- ing straight. Figure 3.16: Distributions of the mean absolute roll rate. The distributions of the mean absolute roll rate are shown in Figure 3.16. The median absolute roll rate is significantly lower for the e-scooter and the e-bike than the bike during the high speed segment. Another result that is shown is that the roll rate is significantly lower when riding straight in a low speed than in a high 40 3. Results speed. For the slalom segment there is no significant difference of the median roll rate between the vehicles but the results indicate that it is higher for the bike. 0.75 0.8 0.85 0.9 0.95 1 R 2 -V a lu e ( 0 -1 ) E-Bike Bike E-Scooter (a) Linear fit between roll rate and steer- ing rate. 0 0.1 0.2 0.3 T im e D e la y ( s ) E-Bike Bike E-Scooter (b) Time delay between roll rate and steering rate. Figure 3.17: Distributions of the correlations between roll rate and steering rate in terms of linear fit and time delay. The three vehicles have a similar linear fit between roll rate and steering rate during slalom since they have similar R2-values which is shown in Figure 3.17a. The median time delay between the roll rate and the steering rate is significantly lower for the e-scooter than the e-bike and the bike which is shown in Figure 3.17b. 3.2.4.2 Segway High Speed Low Speed Slalom 0 4 8 12 16 P it c h A n g le ( d e g ) (a) Average absolute pitch angle at high speed, low speed and slalom with the Seg- way. High Speed Low Speed Slalom 0 2 4 6 P it c h R a te ( d e g /s ) (b) Average absolute pitch rate at high speed, low speed and slalom with Segway. Figure 3.18: Average absolute pitch angle and pitch rate at high speed, low speed and slalom with the Segway. 41 3. Results The pitch angle of the Segway is significantly lower when riding at low speed than at high speed, which is shown in Figure 3.18a. The average absolute pitch rate was very different between the participants at the high speed segment, which can be seen in Figure 3.18b. The results from the figure also indicates that the pitch rate is lower when going straight at low speed than at high speed and slalom. Figure 3.19: Distribution of the mean absolute stick inclination rate with the Segway during slalom. A histogram of the mean absolute stick inclination rate for the Segway is shown in Figure 3.19, where the interval is between 0.5 and 1.5 degrees per second with the most participants between 1.1 and 1.3 degrees per second. 3.3 Questionnaire The mean of the questionnaire results from the 11 different scenarios for each vehicle can be seen in the spider plot in Figure 3.20. It can be seen that the Segway gener- ally is rated the lowest on the different scenarios. The e-bike closely followed by the bike is generally rated the highest in the different scenarios. There are, however, a few questions that differ from this. 42 3. Results 1.0 3.0 5.0 7.0 Mounting and dismounting Accelerating from standing still Turning Maintaining low speed Maintaining high speed Keeping balance at low speedKeeping balance at high speed Braking at low speed Braking at high speed Steering at low speed Steering at high speed E-Bike Bike E-Scooter Segway Figure 3.20: A spider plot of the result of the questions about how the vehicles performed when riding. The scale is converted from very poor, poor, fair, good, very good, excellent and exceptional to 1-7 where 1 corresponds to very poor and 7 to exceptional. 3.3.1 Mounting and Dismounting The lower score on the Segway is reflected in the comments given by seven of the participants saying they were having trouble mounting and dismounting the vehicle. There were two participants commenting on the mounting and dismounting of the bike and e-bike. One of the participants was 164 cm tall and commented that the bike was a bit too large for them, which resulted in troubles getting on and off. The other participant said that they were used to a bike where they could turn the pedals backwards to get a good starting position of the pedals. This is not possible on the bike used in this experiment because of the coaster brake. 3.3.2 Accelerating from Standing Still The accelerating from standing still question resulted in a fairly similar result across the four vehicles with the regular bike being a bit lower. There were some comments about the e-scooter acceleration. One participant commented that they had some problems with the accelerate button. To accelerate with the e-scooter, a small push manually is needed, before it is possible to accelerate with the electric motor. There were three participants who said the e-bike had the best acceleration. There were however one of them saying that the acceleration was a bit too high which made it difficult to keep a steady speed after acceleration. There was another participant 43 3. Results who noted similar difficulties when accelerating with the e-scooter. This other par- ticipant said that it was hard to accelerate gently without sudden changes of speed. 3.3.3 Turning The question regarding turning was not specifically about the experience during the tasks but also the experience the participant had when trying out all the vehicles as well. For this question, it can be seen that the e-scooter is on top and the bike, e-bike and Segway have very similar score. Two comments mentioned that the e-scooter simply was the easiest vehicle to turn with. Another comment mentioned that the Segway had similar turning performance as the e-scooter. Three participants dis- liked turning with the e-bike since they felt the electric assist kick in during a sharp turn resulting in a larger turn radius. 3.3.4 Maintaining Low Speed The only question where the e-bike got the lowest score was the question about maintaining a low speed. There were multiple comments saying that it was hard to maintain a low speed with the e-bike since the motor would kick in a bit too much. Two participants said that they periodically had to brake to not go over the desired speed. It can be seen that the e-scooter also has a lower score at maintaining a low speed which also was reflected in the comments. There were three comments saying the throttle for the e-scooter was very sensitive to small movements. This meant that they always had to do small adjustments on the throttle to maintain a low speed. Contradictory to these three comments were one participant who said that it was easy to keep the throttle in a set position and easily maintain the low speed. 3.3.5 Maintaining High Speed One of the highest average score of the questionnaire is the e-bikes score on main- taining a high speed. The Segway has the lowest score, which is also reflected by multiple participants who felt uncomfortable riding fast with it. There was one par- ticipant who noted that it was exhausting for the legs to ride the Segway fast. The bike had a relatively low score compared to the e-scooter and e-bike. This is also reflected in a comment saying that it was very difficult to maintain a high speed with the regular bike. 3.3.6 Keeping Balance All four vehicles had an average of "Very Good" for how the vehicles perform at keeping balance at low speed. One participant complained about the regular bike being heavy, due to the electric motor and battery, which made it hard to keep the 44 3. Results bike upright. The Segway has the lowest score on keeping balance at high speed followed by the e- scooter then the e-bike and bike. There were two comments saying that balance with the Segway is an experience and practice problem. They said that more experience with the Segway would make it feel more stable and easy to keep balance at high speed. 3.3.7 Braking In both braking at high speed and braking at low speed, the bike and e-bike have the highest and very similar score. The e-scooter and the Segway have similar scores at braking at low speed but when it comes to braking at high speed, the Segway has a lower score than the e-scooter. The e-scooter, compared to the e-bike and bike has a low score. There were participants who complained about the poor brakes on the e-scooter and that it was hard to judge when to initiate the brake. As for the Segway, there were participants who said that if they had more experience riding it they would feel safer with it and be able to brake better. The general score for braking at low speed is higher than braking at high speed. 3.3.8 Steering at Low Speed The Segway has the highest score for steering at low speed and the e-bike has the lowest score. The difference between the scores of all four vehicles are however very small. One of the participants said that while pedaling with the regular bike and steering at the same time resulted in poor balance. Another participants once again said that the heavy weight of the regular bike resulted in bad maneuverability at low speed. 3.3.9 Steering at High Speed Steering at high speed gave scores similar to other questions like the ones about braking and keeping balance. The Segway has the lowest score while the e-bike and bike have similar scores in the top and the e-scooter has a slightly lower score. 3.3.10 Overall The mean of the results of the four questions regarding the overall comfort, maneu- verability, stability and safety can be seen in Figure 3.21. It is clear to see in the graph that the Segway has a lower score on all four overall questions and the e-bike has the highest score in general on the different questions. One can however see that the regular bike is slightly higher rated on the overall safety. The Segway however has a comparable mean score on the overall maneuverability to the other vehicles. 45 3. Results 1.0 3.0 5.0 7.0 Overall Comfort Overall Maneuverability Overall Stability Overall Safety E-Bike Bike E-Scooter Segway Figure 3.21: A spider plot of the overall questions of the vehicles performance. The scale is converted from very poor, poor, fair, good, very good, excellent and exceptional to 1-7 where 1 corresponds to very poor and 7 to exceptional. 46 4 Discussion 4.1 Data Analysis 4.1.1 Speed The mean speed of the vehicles are similar to the study by Violin [11] especially the difference between the vehicles. For the high speed segment the mean speed is generally 1-2 km/h higher in his study which may be since the segmentation was done differently. If the constant segment is started when the participants were still accelerating, the mean speed would be lower. Participants generally had a lower speed with the Segway during the slalom segment which indicate that a lower speed was need to comfortably complete the slalom course. This indicate a lower maneu- verability with the Segway compared to the other vehicle, but could also be due to that the Segway did not have a speedometer. The standard deviation of the speed when the participants were instructed to keep a constant speed is a factor of stability. A low standard deviation indicates that the vehicle is easy to maintain a constant speed with and a high standard deviation indicates that it is hard to keep a constant speed with the vehicle. From the results it could be seen that the Segway is the hardest vehicle to maintain a constant speed with, followed by the e-scooter. The e-bike is harder to maintain a constant speed with than the bike at low speed, but similar at high speed. The reason for this is that the e-bike sometimes gives more power than intended by the user at low speed. The perceived performance by the participants is different to the actual performance when discussing the speed. Like mentioned before, the Segway was rated one of the best vehicles to maintain a low speed with but the results indicate otherwise. The difference may be due to the participants not being able to see their speed while riding the Segway. 4.1.2 Braking The mean deceleration is very similar between the harsh brake and unexpected brake. Since the participants however were prepared that they would do an unex- pected brake, the result could be different compared to a real unexpected situation. The correlation between how often the participants use the bike and the decelera- tion with the bike indicates that more experienced cyclists brake harder. Both when braking gently and harshly. The e-scooter however does not have a clear correlation 47 4. Discussion between braking acceleration and experience. A reason for this may be that even if the participants had a high experience with the e-scooter, it is not sure that they were used to the foot brake. The smaller range however also indicates that the rider has less influence on how fast the vehicle brakes with the e-scooter than the bike and e-bike. The results therefore show that the e-scooter has worse braking capabilities than the e-bike and bike which also corresponds well with the study by Violin [11]. The braking distances for the unexpected brake is also in a similar interval as in the study by Garman et al. [2]. The median braking distance were however lower in that study due to that the participants generally had a lower speed with a mean of 13.7 km/h. That the e-scooter has worse braking capabilities than the e-bike and bike is also indicated by the questionnaire results were braking at high speed were ranked the lowest with e-scooter. The Segway clearly has the lowest deceleration which is a negative aspect in both maneuverability and safety aspects. An alarming result with the e-scooter is that a lot of people passed the stop line before stopping which could be very dangerous in a real setting. Some of these start to brake around the stop line and these participants may have misunderstood or forgotten that they should come to a complete stop before the line and instead believed they should start braking at the line. However since there only was a few cases where the stop line was crossed with the e-bike and the bike, it is fair to say that several participants overestimated the brakes of the e-scooter. This is also the case for some of the riders with the Segway. The braking results are probably affected by which brakes the participants used. Some of the participants used both brakes on the bike, e-bike and e-scooter and some only used one of them. This was not something that was looked at in detail but something that was noticed during the experiments. 4.1.3 Acceleration A high acceleration indicates a high maneuverability. A higher acceleration however also makes it harder to estimate the behavior of the vehicle by other persons in a real setting which is a negative safety aspect. From the acceleration results, it could be seen that the Segway and the e-scooter were the vehicles that could accelerate the fastest in the beginning. It however also has the broadest range of mean accelera- tions which shows that the acceleration behavior differed a lot between participants. All the electric vehicles accelerate faster than the bike, even if the difference between the e-scooter and the bike is small when the participants where told to accelerate fast. The result of the acceleration corresponds well with the results from the study by Violin [11]. From the questionnaire it could be seen that the participants’ perceived performance of the vehicles while accelerating are similar to their actual acceleration performance. The comment regarding the accelerate button on the e-scooter not activating could be the reason for the outlier and long whiskers during the harsh acceleration. 48 4. Discussion 4.1.4 Steering and Leaning The steering angle was generally lower when riding with the e-scooter than with the bike and the e-bike. For the slalom task, this may indicate that the maneu- verability is higher for the e-scooter. For the straight maneuvers, it may indicate a higher stability. Another reason for the lower steering angle could be the smaller wheel base of the e-scooter compared to the bike. The steering angle for the e-bike, bike, and e-scooter are generally lower than in the study by Violin [11]. This may be due to that the segmentation was done differently. The mean absolute steering angle will be considerably higher if a turn or part of a turn is captured. It has also been shown in an earlier study by Kovácsová et al. that the mean absolute steering angle for the bike and the e-bike is considerably higher during acceleration than during straight riding at a constant speed [12]. From the study by Kovácsová et al., it has been shown that the steering angle decreases when the speed increases for e-bikes and bikes [12]. In the study the average mean absolute steering angle were 2.5 degrees when the average mean speed were 7.6 km/h and 1.8 degrees when the average mean speed were 13 km/h for the middle aged participants with the bike. This corresponds well with the results from the bike in this study where the median mean absolute steering angle were 2.0 degrees, when the median mean speed was 10 km/h with the bike. The results were similar for the e-bike. From the results in this study it could be seen that the steering angle also decreases for the e-scooter when the speed increases. The steering rate is lower on the e-scooter which also indicates a higher stability. It is however worth to take into consideration that the pedaling on the e-bike and the bike lead to a motion that affects both the roll motion and the steering motion. This could be a reason that the bike has a significantly higher roll rate and steering rate than the e-scooter and the e-bike at high speed. The stick inclination rate for the Segway is hard to compare with the other vehicles since the vehicles are controlled in such different ways. From the questionnaire, how- ever, it can be seen that the participants rated the Segway the highest for steering at low speeds which would indicate that the Segway is very maneuverable at low speeds. For steering at high speed the Segway was rated relatively low indicating low levels of maneuverability during high speeds. The perceived steering capabilities of the e-bike, bike and e-scooter were very simi- lar. This indicates that the participants feel that the bike was just as stable as the other two even though there is a small difference in the actual results. The peak cross correlation between the steering rate and the roll rate is very high for the three segments. The cross correlation was only calculated at the slalom task where the speed was low. In a study by Cain, it was shown that the cross corre- lation generally is high for both competitive cyclists and regular cyclists at a low speed [20]. This also holds for the e-bike and the e-scooter according to the results from this experiment. The cross correlation values in this study are however much higher than the results from Violin [11] and Kovácsová et al. [12]. In the study by Violin [11] this could depend on how the filtering is done on the steering rate. If the signal contains a lot of noise the correlation will be lower. In this study the peak 49 4. Discussion cross-correlation is estimated during the slalom segment, while it is measured during straight riding in the study by Kovácsová et al. [12]. During the slalom segment, the steering rate and roll rate comes from the turning of the slalom course. When the rider is riding straight, the steering rate and roll rate are inputs done by the rider to stabilize the vehicle. This may be a reason that the peak cross correlation is different. The time delay between the steering rate and roll rate is lower for the e-scooter which indicates a higher maneuverability at a low speed than the bike and the e-bike. The time delay of the bike and the e-bike are similar to the study by Kovácsová et al. [12]. The time delay for the e-bike, bike and e-scooter are similar compared to the earlier study by Violin [11]. Another aspect of steering and leaning is keeping balance which is directly connected to stability. From the questionnaire the participant rated the four vehicles similarly to the questions about steering. The Segway is rated similar at keeping balance at high and low speeds where the other three vehicles were rated higher at keeping balance at high speed than low speed. The reason why the bike and e-bike is rated higher on the high speed is probably due to that a bike is self-stable at high speeds [1]. 4.1.5 Mounting and Dismounting Troubles mounting and dismounting in real traffic could cause problems when, for example, stopping and starting from a traffic light. By not being able to dismount the vehicle after coming to a stop could result in a fall. Not being able to mount the vehicle again after coming to a stop could result in delaying traffic. Mounting and dismounting could be seen as both stability and comfort aspect. If it is difficult to mount and dismount a vehicle it might be because the vehicle is unstable. During the experiments and try-out session, nearly all of the participants needed help with getting on and off the Segway. This together with the questionnaire results indicates that the Segway has the lowest levels of stability and comfort during mounting and dismounting. 4.2 Limitations 4.2.1 Data Sets Availability During the testing there were some problems that led to missing or corrupted data. The reason for missing LIDAR data was caused by either the 12