Scooter: Multifaceted Research Perspectives

• Scooter is a versatile research object defined by its roles in combustion optimization, battery-electric design, shared mobility operations, participatory sensing, and adversarial machine learning evaluation.
• Methodologies span SOCP-based optimization, wireless communication measurements, social media data analytics, and human-in-loop experimental simulations across diverse platforms.
• Implications include improved fuel economy, optimized urban fleet operations, enhanced safety protocols, privacy-preserving data strategies, and innovative VR/HCI interfaces.

Scooter denotes several distinct but technically connected objects in contemporary research: a restricted combustion-powered 50 cc vehicle optimized through engine and throttle control, a battery-electric micromobility vehicle, a shared dockless fleet asset embedded in urban transport systems, a sensorized research platform, and, in one machine-learning context, an acronymic evaluation framework. Across these usages, the research literature treats scooters as coupled mechanical, cyber-physical, infrastructural, and policy objects whose performance depends on propulsion architecture, human behavior, road geometry, fleet operations, and data governance (Kreß et al., 2024, Korzilius et al., 2021, Feng et al., 2020, Fazlija et al., 10 Jul 2025).

1. Research scope and typology

Recent arXiv literature uses the term across several non-identical domains. In vehicle engineering, a scooter may be a Peugeot Kisbee 50 4T (Euro 5) with combustion control and exhaust optimization (Kreß et al., 2024). In micromobility research, it usually denotes the shared dockless e-scooter or the privately operated battery electric micromobility vehicle studied in urban travel, safety, and fleet design (Feng et al., 2020, Korzilius et al., 2021). In operations and autonomy, scooters appear as self-repositioning or autonomous shared assets (Kondor et al., 2019, Tan et al., 5 Oct 2025). In research infrastructure, they function as participatory sensing platforms (Khan et al., 10 Jan 2025). In virtual-environment work, scooter riding is abstracted into locomotion and simulation metaphors (Zhou et al., 12 Apr 2026, He et al., 5 Jan 2026). Separately, SCOOTER is expanded as Systemizing Confusion Over Observations To Evaluate Realness, a framework for human evaluation of unrestricted adversarial examples (Fazlija et al., 10 Jul 2025).

Research context	Scooter formulation	Representative source
Combustion vehicle engineering	50 cc Euro 5 scooter with velocity-controlled Throttle-by-Wire	(Kreß et al., 2024)
Electric micromobility design	Battery, single electric motor, fixed-gear transmission, final drive, driven wheel	(Korzilius et al., 2021)
Shared urban mobility	Dockless e-scooter service with rider, gig-worker, and operator roles	(Feng et al., 2020)
Fleet autonomy	Self-repositioning shared personal mobility device or ASMV	(Kondor et al., 2019, Tan et al., 5 Oct 2025)
Sensing testbed	Retrofitted e-scooter with WBSC, FC, and RAMP	(Khan et al., 10 Jan 2025)
VR / HCI	Handlebar-based virtual locomotion or modular micromobility simulator mode	(He et al., 5 Jan 2026, Zhou et al., 12 Apr 2026)

This multiplicity is not merely terminological. It indicates that scooter research spans internal-combustion calibration, battery sizing, communication channels, rider cognition, multimodal interaction, transit substitution, privacy-preserving governance, and human-in-the-loop evaluation.

2. Vehicle engineering, control, and communication

In combustion-engine research, scooters are studied as tightly constrained thermodynamic systems. A 2024 study on modern 50 cc Euro 5 scooters replaced the usual restriction strategy—described as speed reduction through negatively shifted ignition timing—with a velocity-controlled Throttle-by-Wire system (TbWS) that regulates throttle valve opening instead while maintaining $\lambda = 1$. The platform was a Peugeot Kisbee 50 4T (Euro 5) tested on a roller dynamometer after a coast down test, with engine signals including injection quantity, engine speed, ignition timing, cylinder wall temperature, exhaust gas temperature, oxygen sensor data, crankshaft position, and in-cylinder pressure, plus CAN variables such as throttle opening angle, rider acceleration command, and vehicle velocity. Exhaust instrumentation measured CO, CO2, NOx, O2, and HC, as well as temperature and mass flow. On level ground, a difference of 50% in the throttle opening yielded a 17% improvement in fuel economy and 17% reduction in exhaust gas flow; CO emissions decreased by a factor of 8.4, CO2 by 1.17, and HC by 2.1, while NOx increased by a factor of 3 (Kreß et al., 2024). The same paper gives the exhaust-flow estimate

$$m_{exh}= (Fuel_{con}\cdot \rho_{fuel})\cdot (1 + 14.7\lambda).$$

The reported increase in internal cylinder pressure under the optimized strategy was used to substantiate improved combustion phasing (Kreß et al., 2024).

In electric-vehicle design, the scooter is modeled as a battery-electric micromobility vehicle with Battery (BAT), Single electric motor (EM), Fixed-gear transmission (FGT), Final drive (FD), and Driven wheel (W). For the e-scooter case, the study assumes $v_{\max}=25\ \mathrm{km/h}$, driver mass $m_d=75\ \mathrm{kg}$, frame mass $m_f=10\ \mathrm{kg}$, wheel radius $r_w=0.125\ \mathrm{m}$, and auxiliary power $P_{aux}=10\ \mathrm{W}$, and frames an optimal joint design-and-control problem minimizing

$$J_{\mathrm{TCO}}=C_{\mathrm{op}}+C_{\mathrm{comp}}.$$

Because the full problem is nonlinear in motor size and total mass but convex for fixed values, the authors solve it via an SOCP-based iterative method. The reported flat-terrain optimum is $P_{em,\max}=590\ \mathrm{W}$, $E_{b,\max}=435\ \mathrm{Wh}$, $$m_{exh}= (Fuel_{con}\cdot \rho_{fuel})\cdot (1 + 14.7\lambda).$$0, $$m_{exh}= (Fuel_{con}\cdot \rho_{fuel})\cdot (1 + 14.7\lambda).$$1, and $$m_{exh}= (Fuel_{con}\cdot \rho_{fuel})\cdot (1 + 14.7\lambda).$$2. In the hilly scenario, the optimum becomes $$m_{exh}= (Fuel_{con}\cdot \rho_{fuel})\cdot (1 + 14.7\lambda).$$3, $$m_{exh}= (Fuel_{con}\cdot \rho_{fuel})\cdot (1 + 14.7\lambda).$$4, $$m_{exh}= (Fuel_{con}\cdot \rho_{fuel})\cdot (1 + 14.7\lambda).$$5, $$m_{exh}= (Fuel_{con}\cdot \rho_{fuel})\cdot (1 + 14.7\lambda).$$6, and $$m_{exh}= (Fuel_{con}\cdot \rho_{fuel})\cdot (1 + 14.7\lambda).$$7. The same study concludes that regenerative braking and gear-changing capabilities may not be worth implementing for the use cases examined (Korzilius et al., 2021). This suggests that, in lightweight micromobility, component-cost and mass penalties can dominate theoretical drivetrain-efficiency gains.

Scooter engineering also includes wireless communication. A scooter-to-X communication study using a YAMAHA Cygnus-X 125 (2011 model) with IEEE 802.15.4 / ZigBee at 2.480 GHz found that human body shadowing introduces scooter-specific propagation losses. Among the tested antenna placements, the left mirror was the best practical location, but the driver alone still caused about 9–12 dB extra attenuation, and driver plus passenger produced up to 17.9 dB additional attenuation; the paper summarizes the overall effect as 9–18 dB average attenuation. In system simulations with 70% scooters / 30% cars, including body shadowing reduced the average number of received packets by as much as 40% relative to models that ignored the effect (Lin et al., 2015). A plausible implication is that scooter networking cannot be treated as a simple down-scaled version of car-to-car communication.

3. Shared micromobility, demand, and fleet operations

Shared scooter systems are studied as urban transport networks with heterogeneous mode interactions. A large social-data study collected 5.8 million scooter-tagged tweets and 144,197 images from 2.7 million users between October 2018 and March 2020, later filtering to 416,291 tweets and 17,695 images about shared dockless e-scooters. It identified four broad discussion themes—deployment, stakeholders, operations, and emotion—and showed that the dominant brands in both text and image analysis were Lime, Bird, and Lyft. Among analyzed rider images, 83.51% were not wearing helmets; among 230 unduplicated parking images, only 37.39% showed proper parking, while 62.61% were in wrong places, including 34.78% of all scooters parked in the middle of sidewalks (Feng et al., 2020). The same study reported a median payment of \$3.8 and median trip duration of 15.0 minutes from extracted app screenshots, and found that positive expressions in the word-based sentiment analysis were 3.48 times as frequent as negative ones (Feng et al., 2020). These observations position scooters simultaneously as accepted transport options and recurrent sources of operational conflict.

Demand forecasting studies treat scooters as substitutes for some modes and complements to others. A Manhattan-focused model estimated that a deployment of 2,000 scooters would produce 75,000 e-scooter trips per day, about 1% of total Manhattan trips, corresponding to approximately \$m_{exh}= (Fuel_{con}\cdot \rho_{fuel})\cdot (1 + 14.7\lambda).$$877 million annual revenue</strong> under the fare assumption reported in the study. The same model estimated that e-scooters could replace <strong>32 <p>Fleet-rebalancing research extends the scooter from a passive rental asset to an autonomous logistics agent. A shareability-network study on <strong>self-repositioning shared personal mobility devices (SRSPMDs)</strong>, using dockless bike-share data in Singapore as a demand proxy, reported an ideal fleet of about <strong>4,000 vehicles at$$m_{exh}= (Fuel_{con}\cdot \rho_{fuel})\cdot (1 + 14.7\lambda).$9 km/h and 1,500–2,000 vehicles at $v_{\max}=25\ \mathrm{km/h}$0 km/h, compared with 13,500–18,000 active bikes per day in the real system, yielding up to 10x higher utilization than current bike-share systems (Kondor et al., 2019). More recently, the SMART framework incorporated a small number of autonomous shared micromobility vehicles (ASMVs) into Chicago e-scooter operations. With only 3% ASMVs, reported demand-satisfaction rates improved from 68.95% to 89.84% for SDSM, 82.05% to 92.39% for GA, and 90.61% to 97.46% for RECOMMEND (Tan et al., 5 Oct 2025). This suggests that, in operational terms, a scooter fleet can be optimized not only by bulk redistribution but also by limited autonomous adaptation.

4. Infrastructure, safety, and interaction with other road users

Safety research consistently treats scooters as infrastructure-sensitive vehicles. A naturalistic multimodal field study with 23 participants instrumented with eye-tracking glasses, cameras, and a bike computer compared three infrastructures: a pedestrian-shared path, a cycle lane, and a roadway. Reported mean speeds were 16.70 km/h on the pedestrian-shared path, 18.14 km/h in the cycle lane, and 17.11 km/h on the roadway. Across all participants, 371 speed change points were detected, with the roadway showing the highest average number and the cycle lane the fewest. The authors concluded that the cycle lane showed the highest average speed, the least speed change points, and the least head movements, supporting its suitability as dedicated infrastructure for e-scooters (Kegalle et al., 24 Feb 2025). The same study highlighted a design and regulatory issue: hand signaling may be risky for scooter riders because taking one hand off the handlebars can compromise stability (Kegalle et al., 24 Feb 2025).

A separate naturalistic gaze study in Charlottesville used a Ninebot MAX KickScooter, Tobii Pro Glasses 3, and GPS/speed logging over 16 trips totaling about 16 hours. It found that bike lanes yielded more focused and stable gaze behavior, whereas roads without bike lanes induced higher gaze variability, higher stationary gaze entropy (SGE) and gaze transition entropy (GTE), and more frequent fixation. The paper formalized

$v_{\max}=25\ \mathrm{km/h}$1

Among listed traffic scenarios, “Bike lane to cross.” had the highest SGE at 5.77 and GTE at 28.68, while “Downhill” and “Road to B.L.” had the highest PRC at 0.72 (Chen et al., 2024). This suggests that transition zones and mixed-traffic geometries impose distinct attentional loads not captured by coarse crash statistics alone.

Pedestrian-facing safety work shows that scooter risk is also spatio-temporally clustered. A crowd-sensed field study on two UTSA campuses recruited 105 participants, of whom 77 completed all assigned tasks and were analyzed. The study detected 1800 predicted encounters from BLE sensing and analyzed 4993 feedbacks between 06:00 and 23:00. It found that about 20% of recorded observations in that window corresponded to moving scooters, and at least 100 observations involved scooters approaching pedestrians from behind. In almost 60% of moving encounters, participants exhibited elevated heart rate when scooters came within about one foot from the front or behind. Spatially, sidewalks and local streets had the highest encounter burdens, and more than 90% of spatio-temporal zones had no encounters, indicating that risk was highly concentrated rather than uniform (Maiti et al., 2019).

Interaction with vehicles has also been modeled explicitly. A vehicle–electric scooter interaction (VEI) simulator represented the e-scooter with a point-mass Newtonian dynamic, a social force model, and a finite state machine (FSM) under a geometric field-of-view perception model. For Aggressive scooter behavior, reported collision rates were 22.22% in the one-vehicle crossing scenario, 36.75% in the two-vehicle crossing scenario, and 18.67% in the lane-changing scenario; Normal scooters had 0.00% collision rate in all three scenarios under the reported settings (He et al., 2024). Complementing this, an e-Scooter Collision Avoidance System (eCAS) combined SR-LSTM pedestrian trajectory prediction with artificial potential field (APF) planning, using

$v_{\max}=25\ \mathrm{km/h}$2

and reported MAD = 0.71 m and FAD = 1.46 m on ETH and UCY (Yan et al., 2023). Together these works indicate that scooter safety research now spans empirical infrastructure studies, human physiological response, interaction simulation, and predictive motion planning.

5. Data infrastructures, privacy, and system observability

Scooter research increasingly depends on purpose-built sensing infrastructures. ScooterLab is a participatory sensing testbed built from retrofitted battery-powered micromobility vehicles. Its scooter-level unit, the Wireless Base Station Computer (WBSC), uses a Raspberry Pi 4 (4GB) connected to a Sense HAT V2, an Adafruit Ultimate GPS module, a Pi Camera Module 3, and a USB microphone, and is powered from the scooter battery through a 40V to 5V DC/DC step-down converter. The initial fleet comprises eight Segway G30 Max scooters with range up to 40 miles and maximum speed 18 mph. Data are uploaded to the Fleet Controller (FC) when the scooter comes within range of the UTSA Wi‑Fi network, while the Research Activities Management Portal (RAMP) supports project setup, a Map tool based on the ArcGIS Maps SDK for JavaScript, and a Stats tool for tabular and chart-based exploration (Khan et al., 10 Jan 2025). This provides a concrete architecture for instrumenting scooters as moving urban observatories.

At the same time, scooters have become a canonical example in mobility-data privacy debates. A study of Los Angeles argued that the city’s proposed Mobility Data Specification (MDS) was unnecessary for several stated oversight use cases because GBFS already supports fleet-size estimation and neighborhood-distribution analysis. However, the same paper showed that even free_bike_status.json data on parked scooters can enable trip reconstruction and inference of sensitive destinations. To mitigate that risk, it proposed a geo-indistinguishability mechanism satisfying

$v_{\max}=25\ \mathrm{km/h}$3

with polar Laplacian density

$v_{\max}=25\ \mathrm{km/h}$4

For a representative setting of $v_{\max}=25\ \mathrm{km/h}$5 km and $v_{\max}=25\ \mathrm{km/h}$6, the study reported about 115 scooters lost from LA boundaries, roughly 3%, and about 4 scooters lost per neighborhood on average (Baltra et al., 2020). The result is not that scooter data are unusable, but that utility claims and privacy claims must be jointly quantified.

Battery observability is another systems problem. A stochastic model for electric scooter systems, motivated by 71,518 likely trips reconstructed from JUMP data in Washington, D.C., tracked the fraction of scooters in battery-life buckets rather than each scooter individually: $v_{\max}=25\ \mathrm{km/h}$7 The empirical analysis reported median trip duration around 8 minutes, median distance around 800 meters, and median battery use per trip around 6%, with about 80% of trips using less than 10% battery. The paper then proved a mean field limit theorem and a functional central limit theorem, and used the asymptotic approximations to size battery-swapping staff. For the target $v_{\max}=25\ \mathrm{km/h}$8, the staffing algorithm yielded $v_{\max}=25\ \mathrm{km/h}$9 (Pender et al., 2020). A plausible implication is that scooter system observability increasingly requires stochastic population-level models rather than per-vehicle heuristics.

6. Simulation, virtual environments, and terminological extension

Scooters also function as embodied interfaces in VR and HCI. MicroVRide is a modular 4-in-1 VR micromobility simulator supporting e-scooters, Segways, electric unicycles, and one-wheeled skateboards on a single platform. In scooter mode, the platform remains fixed, a handlebar is attached, a handlebar-mounted IMU maps yaw $m_d=75\ \mathrm{kg}$0 steering, and a thumb throttle controls velocity. Reconfiguration between vehicles takes about one minute. In a preliminary within-subject study with $m_d=75\ \mathrm{kg}$1, the scooter condition had the lowest raw NASA-TLX workload, $m_d=75\ \mathrm{kg}$2, and was consistently described as the easiest, most natural, and most realistic of the four vehicles (Zhou et al., 12 Apr 2026).

A related system, LocoScooter, is a stationary scooter-based locomotion interface for VR navigation in confined spaces. It combines one foot sliding on a compact treadmill-like base with handlebar yaw control, occupies a 0.5 m² footprint, and costs about \$230 in commodity hardware. In a within-subject study with $m_d=75\ \mathrm{kg}$3, LocoScooter did not differ significantly from joystick navigation in task completion time—$m_d=75\ \mathrm{kg}$4 s for LocoScooter versus $m_d=75\ \mathrm{kg}$5 s for joystick, $m_d=75\ \mathrm{kg}$6—but significantly improved hedonic quality, overall user experience, involvement, realism, and enjoyment, while increasing physical demand without increasing reported fatigue (He et al., 5 Jan 2026). These results treat scooter riding not as transport but as a sensorimotor metaphor for controlled virtual movement.

The term also extends outside transportation. In adversarial machine learning, SCOOTER—Systemizing Confusion Over Observations To Evaluate Realness—is an open-source framework for human evaluation of unrestricted adversarial examples. It organizes evaluation into preliminary screening, main Likert-scale annotation, and equivalence-based statistical testing using TOST with bounds $m_d=75\ \mathrm{kg}$7 and $m_d=75\ \mathrm{kg}$8. Across 346 human participants, over 34K human ratings, 3K real images, and 7K adversarial examples, the reported result was that three color-space attacks and three diffusion-based attacks failed to produce imperceptible images under the framework’s criterion (Fazlija et al., 10 Jul 2025). The reuse of the term for a human-evaluation protocol is terminologically separate from vehicle research, but it shows that “SCOOTER” has become a recognizable label for structured experimental methodology as well as for mobility technology.

Taken together, these literatures depict the scooter as a dense research object rather than a single vehicle category. It is simultaneously an emissions-constrained engine platform, a battery-electric design problem, a shared fleet unit, a vulnerable-road-user interface, a data source, a privacy risk surface, a VR control metaphor, and, in another field, a human-evaluation framework. The strongest cross-paper regularity is not a single performance number or policy conclusion, but the repeated finding that scooter outcomes are highly contingent on control strategy, infrastructure, sensing, and context (Kreß et al., 2024, Kegalle et al., 24 Feb 2025, Baltra et al., 2020, Fazlija et al., 10 Jul 2025).