Road-to-Traveler-Feedback Device

• Road-to-Traveler-Feedback Devices are traveler-facing systems that convert sensed road-state data into clear, actionable guidance.
• They integrate distributed sensing, state estimation, and decision logic to provide personalized, real-time feedback across diverse transportation modalities.
• Architectural designs range from centralized to crowd-sourced models, with studies addressing interface design, latency challenges, and multimodal communication.

A Road-to-Traveler-Feedback Device (RTFD) is a traveler-facing information system that converts sensed, inferred, or precomputed road-state information into actionable guidance for a road user. In the cited literature, the term spans several technical realizations: a smartphone application that advises motorists how fast to travel to remain inside a “Ride-the-Green-Wave” on signalized arterials (Friedman et al., 2024); a vehicle-mounted external human–machine interface (eHMI) that communicates autonomous-vehicle intent to pedestrians (Schmidt-Wolf et al., 2022); a stationary Android-based crossing aid that classifies unsignalized crossings into TRAFFIC and GAP states (Perry et al., 2016); a city-scale accessibility feedback system for people with disabilities (Polenakis et al., 2024); and real-time passenger information, routing, and in-car augmented-reality systems that expose network state, route quality, or points of interest to travelers (Ganesh et al., 2012, El-Wakeel et al., 2018, Schramm et al., 12 Feb 2025).

1. Conceptual scope

Across these works, the device is best understood as an interface layer between traffic or mobility state and traveler action. In autonomous-vehicle pedestrian interaction, it is explicitly defined as an external HMI mounted on the vehicle that visually communicates the vehicle’s current or imminent behavior and the recommended pedestrian action (Schmidt-Wolf et al., 2022). In accessibility assistance, the same idea expands into the front end of a larger distributed cyber-physical system that monitors ramps and crossings and feeds live accessibility status back to travelers and authorities (Polenakis et al., 2024). In green-wave control, the RTFD becomes the practical mechanism that tells a driver the recommended speed, their position relative to a moving green wave, and whether they are ahead of, behind, or inside that wave (Friedman et al., 2024, Friedman et al., 16 Aug 2025).

Form	Primary traveler	Representative function
Vehicle-mounted eHMI	Pedestrian or nearby passenger	Communicate yielding, non-yielding, or crossing invitation
Roadside crossing aid	Pedestrian at unsignalized crossing	Indicate TRAFFIC or GAP
Mobile or city-scale feedback client	Person with disabilities	Report blocked or semi-blocked ramps and reroute
Passenger information system	Public-transport rider	Show ETA, arrivals, and live vehicle positions
Routing or progression device	Driver	Recommend route, segment quality, or green-wave speed
In-car AR interface	Passenger	Surface passed and upcoming points of interest

This variety suggests that the “device” is not restricted to a single hardware class. It can be a smartphone, an in-vehicle display, a vehicle exterior display, a roadside unit, or the front end of a cloud-backed mobility service. What remains invariant is the functional role: the system senses or models conditions on the road, interprets them with respect to traveler goals, and returns a concise recommendation, warning, or explanation.

2. Architectural patterns

The dominant architectural pattern is distributed sensing plus centralized or semi-centralized interpretation. The accessibility system for people with disabilities is explicitly modular, with the Mark Inaccessible City Point System, the Live Data Analysis and Response System, and the Obstruction Detection and Prevention System linked in a client–server design. Sensor nodes at ramps and mobile clients both send events to a central server, which updates accessibility states and pushes traveler feedback while also notifying authorities (Polenakis et al., 2024). The real-time passenger information system follows the same structure: vehicle-mounted units report GPS coordinates via GPRS, station units fetch ETA values, and a central server maintains route, link, stop, and bus-state tables while running a link updater and ETA calculator (Ganesh et al., 2012).

Road-quality-aware routing and crowd-based route recommendation place the computational burden in cloud services. In iDriveSense, vehicles and smartphones contribute geo-referenced anomaly data to Road Information Services, which first assess road segments and then evaluate candidate routes using two cascaded fuzzy inference systems (El-Wakeel et al., 2018). CrowdPlanner similarly separates candidate generation from route verification: a traditional route recommendation module proposes candidate routes, and a crowd-based module generates landmark questions and selects workers to determine the best route when algorithmic confidence is insufficient (Su, 2013). By contrast, the mobile-observer method is comparatively edge-centric: each communicating vehicle can act as a moving traffic sensor using only received beacon messages, although aggregation on a server is the natural extension for network-wide feedback (Phu, 2023).

In immersive passenger interfaces, the architecture adds precise tracking and rendering. The in-car AR system combines a video see-through head-mounted display, eye tracking, hand tracking, and vehicle-mounted inertial sensing to keep points of interest world-fixed while presenting seat-fixed List, Timeline, and Minimap views (Schramm et al., 12 Feb 2025). In adaptive longitudinal control, the architecture shifts again: the “feedback device” is the driving function itself, which logs driver interventions during traded control and later adjusts its route-specific speed profile (Schwager et al., 14 Feb 2026). The common pattern is thus not a single topology but a repeated decomposition into sensing, state estimation, decision logic, and traveler-facing feedback.

3. Sensing, estimation, and formal models

The mathematical core of RTFD systems varies by domain but repeatedly takes the form of reducing rich physical observations to a compact traveler-relevant state. At unsignalized pedestrian crossings, optical flow from a roadside Android camera is projected onto an influx map to form a one-dimensional Activity signal,

$$A_n = \sum_{x,y} \vec{m}(x,y) \cdot \vec{u}_n(x,y),$$

and a matched-filter likelihood-ratio test classifies a sliding window as TRAFFIC or GAP (Perry et al., 2016). This produces a binary crossing recommendation rather than a full scene interpretation, which is consistent with the paper’s emphasis on simple, early warning.

In traffic estimation from communicating vehicles, the mobile-observer method derives counts of same-direction and opposite-direction encounters from beacon durations and relative speeds, then estimates flow and density. A central relation is

$q = \frac{x_1 + x_2}{t_1 + t_2},$

with further estimators for density and penetration derived from binomial likelihoods and contact-duration models (Phu, 2023). The method is notable for requiring no radar or GPS in the core estimator.

Routing systems replace physical-state detection with graph costs and fuzzy evaluations. The accessibility framework models the city road network as $G=(V,E)$ and assigns each transition point an accessibility state $a(v,t)\in\{0,1,2\}$, which then affects routing (Polenakis et al., 2024). iDriveSense computes road-segment quality from anomaly percentage, severity, and lane width using a Mamdani fuzzy inference system, then evaluates candidate routes with a second fuzzy model that trades off average segment quality, time, and distance (El-Wakeel et al., 2018). CrowdPlanner formalizes candidate routes as landmark-based routes and selects discriminative landmark sets that maximize significance while minimizing question burden (Su, 2013).

Signal-progression systems use explicit timing geometry. On Ride-the-Green-Wave roads, green-wave speed is defined by

$v_g = \frac{D}{T_g},$

where $D$ is the distance between adjacent RGW-nodes and $T_g$ is the forward green duration (Friedman et al., 16 Aug 2025). The RTFD then maps the driver’s position and time within the cycle to a recommended speed and relative position inside the moving green wave (Friedman et al., 2024). In adaptive longitudinal driving, personalization is expressed as iterative speed-profile adjustment over route distance $d$, with the updated profile combining the baseline predictive profile and a preprocessed intervention trajectory,

$$v_{\text{mean}}(d)=\frac{v_{\text{PLDF}}(d)+v_{\text{prepro}}(d)}{2},$$

before smoothing and constraint enforcement (Schwager et al., 14 Feb 2026).

4. Communication modalities and interface design

The most extensively studied traveler-facing modality in this literature is external communication between vehicles and pedestrians. A vehicle-to-pedestrian study evaluates four external display classes—text, symbols, light-based displays, and road projections—and finds that symbols rank highest overall for preference, legibility, trust, and public acceptance, while road projections rank worst (Schmidt-Wolf et al., 2022). For yielding situations, the preferred design is the text “Safe to cross” combined with a walking-person symbol; “Walk” is acceptable, but lights-only and projection-only interfaces are treated as more ambiguous (Schmidt-Wolf et al., 2022). The study’s statistical treatment is explicitly ordinal, using Shapiro–Wilk, Kruskal–Wallis, and Mann–Whitney U tests.

Passenger exposure changes the design problem. In autonomous personal mobility vehicles, silent text-only eHMI creates a “silent time” in which the vehicle communicates to pedestrians but not to the onboard passenger. In a field study, this makes eHMI-T less suitable than the multimodal alternatives. Neutral-voice multimodal eHMI has an advantage in pragmatic quality, while affective-voice multimodal eHMI has an advantage in hedonic quality (Liu et al., 2023). The study further reports that more introverted or less open passengers tend to prefer simpler communication, whereas more extraverted, open, and agreeable passengers tend to prefer affective voice (Liu et al., 2023). This suggests that personality-adaptive external communication is not merely cosmetic but may alter passenger comfort.

At roadside crossings, feedback modalities diversify further. The Android-based safety device can indicate crossing state by sound, display, vibration, or local communication such as Bluetooth or WLAN, reflecting an explicitly multimodal accessibility orientation (Perry et al., 2016). In in-car AR, the traveler is not a pedestrian or driver but a passenger exploring world-fixed points of interest. There, the principal modality question is not text versus symbol but List versus Timeline versus Minimap, combined with eye-gaze and pinch. The field study finds general acceptance, with the List visualization preferred, and identifies the impact of vehicle movement on three-dimensional interaction as a central hardware limitation (Schramm et al., 12 Feb 2025).

5. Application domains

One major application domain is local crossing negotiation. Here, the RTFD acts at the scale of seconds and meters, helping a pedestrian decide whether to cross or helping an autonomous vehicle communicate whether it will yield. Representative systems include vehicle-front eHMIs, APMV external communication, and stationary roadside crossing aids (Schmidt-Wolf et al., 2022, Liu et al., 2023, Perry et al., 2016). The output is usually categorical and immediate: safe to cross, do not cross, yielding, thanks, or gap available.

A second domain is accessibility and assisted mobility. The distributed feedback system for people with disabilities treats ramps and crossings as monitored transition points whose accessibility status can be accessible, semi-blocked, or fully blocked. The traveler-facing device is the mobile application that displays live map status, alerts on-route obstructions, and recomputes routes (Polenakis et al., 2024). This is still a road-to-traveler feedback device, but its temporal scale is minutes rather than seconds, and its state includes infrastructure obstruction rather than vehicle intent.

A third domain is collective passenger information. The real-time passenger information system uses periodic position reports from buses to estimate arrival times and then presents ETA through stop displays, audio announcements, web interfaces, and SMS (Ganesh et al., 2012). In-car AR extends this collective-information model from transport operations to environmental exploration by letting passengers revisit passed points of interest and inspect upcoming ones (Schramm et al., 12 Feb 2025).

A fourth domain is route and motion optimization. iDriveSense feeds road-surface quality back to drivers as route suggestions (El-Wakeel et al., 2018); CrowdPlanner feeds crowd knowledge back into route choice (Su, 2013); the mobile-observer method can supply live density and flow estimates to routing systems (Phu, 2023). At the most tightly coupled end of this spectrum, the adaptive predictive longitudinal driving function uses driver interventions as implicit feedback to refine future speed profiles, thereby converting the vehicle itself into a route-specific learning feedback device (Schwager et al., 14 Feb 2026). Green-wave systems occupy a special position: they are neither merely informational nor fully autonomous, but prescriptive, advising motorists how to move so that infrastructure timing and traveler motion remain synchronized (Friedman et al., 2024, Friedman et al., 16 Aug 2025).

6. Evaluation, limitations, and future directions

The literature evaluates RTFDs with heterogeneous but rigorous methodologies. Human-facing interface papers rely on Likert ratings, rankings, nonparametric hypothesis tests, UEQ or UEQ-S, SUS, RTLX, MISC, and personality inventories (Schmidt-Wolf et al., 2022, Liu et al., 2023, Schramm et al., 12 Feb 2025). Algorithmic systems report estimator validity against simulation or prototype deployments, such as VEINS for beacon-based traffic estimation, RGW-SIM for Ride-the-Green-Wave timing, or campus-scale demonstrations for real-time passenger information (Phu, 2023, Friedman et al., 2024, Ganesh et al., 2012). The adaptive driving-function study reports significant increases in satisfaction and significant reductions in intervention frequency with the adaptive predictive longitudinal driving function, but also stresses that learning every intervention indiscriminately is undesirable (Schwager et al., 14 Feb 2026).

Several recurring limitations appear. Visual-only or silent interfaces can exclude some affected travelers, as shown by the APMV passenger study (Liu et al., 2023). Abstract light patterns and road projections may be less legible than symbol-based communication (Schmidt-Wolf et al., 2022). Optical-flow crossing aids face weather and night-time challenges (Perry et al., 2016). In-car AR is constrained by current hardware and by the effect of vehicle motion on fine-grained pinch interaction (Schramm et al., 12 Feb 2025). Crowd-based routing introduces latency, worker selection, and truth-reuse considerations (Su, 2013). Mobile-observer estimation depends on penetration ratio, communication reliability, and assumptions about relative speeds and contact duration (Phu, 2023). Green-wave deployment presupposes coordinated cycle times, suitable node placement, and, in some cases, left-turn-around geometry (Friedman et al., 2024, Friedman et al., 16 Aug 2025).

Future work therefore converges on a few technical directions. Multiple papers call for more realistic validation in VR, field trials, and real deployments (Schmidt-Wolf et al., 2022, Polenakis et al., 2024). Several systems point toward tighter smart-city integration, including infrastructure coordination, V2X dissemination, and multimodal assistance for cyclists, pedestrians, and people with disabilities (Polenakis et al., 2024, Friedman et al., 2024). Interface work points toward adaptive personalization, whether through personality-sensitive eHMI voicing or route-specific learning from implicit driver feedback (Liu et al., 2023, Schwager et al., 14 Feb 2026). Taken together, these studies indicate that the Road-to-Traveler-Feedback Device is not a single mature artifact but a convergent design pattern in intelligent transportation systems: one that couples sensing, inference, timing, and human-facing communication to make road environments legible, traversable, and, in some cases, optimizable.