\subsection{Prototype Design} Guided by the findin...

Based on the provided description and the image you have uploaded, here's an updated version of your text with the necessary modifications:

---

Prototype Design

Guided by the findings, we designed a conceptual interaction system for AV-to-pedestrian communication. Our design integrated multiple modalities—visual, auditory, and kinematic cues—to maximize clarity and redundancy. We chose to implement the prototype in a 3D simulation environment (using Carla~\cite{Dosovitskiy17}) for flexibility and safety, allowing us to create realistic crossing scenarios and iterate on interface elements quickly.

\begin{figure}[htbp] \centering \includegraphics[width=\linewidth]{prototype_design.png} % 图片文件名 \caption{Prototype Design of the AV-to-Pedestrian Communication System} % 图片标题 \label{fig:prototype_design} % 图片标签 \end{figure}

Visual Display: The core of the system is an external LED display mounted on the front of the vehicle (just above the grille). During normal driving, the display remains off or shows a neutral pattern. When the vehicle’s sensors detect a pedestrian intending to cross (for example, someone waiting at a crosswalk), the display activates to communicate the vehicle’s state. We devised two main display states: an acknowledgment state and a yielding state.

In the acknowledgment state, the display shows a simple eye-like graphic that “locks onto” the pedestrian (animated to look in the pedestrian’s direction) along with a short text “I see you”. This directly addresses the pedestrian’s primary concern of being recognized. The graphic was kept minimalistic to avoid uncanny valley effects—just two LED oval shapes representing eyes that orient toward the pedestrian’s position.

Following acknowledgment, if the vehicle is going to yield, the display transitions to the yielding state: it shows a walking pedestrian icon (a standard white walk symbol similar to traffic lights) accompanied by the text “Safe to cross”. Both the icon and text are illuminated in a calm cyan color, chosen for its association with permissive signals in some prior concept studies and for being color-blind friendly (distinguishable from red).

If the vehicle needs to continue (for instance, it cannot stop in time or has priority), the display would instead show a stop hand icon (red) or a message like “Stopping not possible” alongside a caution light pattern. However, for our initial prototype and tests, we focused on the yielding scenario, since our use case was an unsignalized crosswalk where the AV is intended to yield to pedestrians.

As shown in Figure~\ref{fig:prototype_design}, the visual display is a crucial component of the communication system. The two main display states are represented clearly to ensure pedestrian safety and clarity in communication.

Auditory Cue: As an optional enhancement, we added a gentle auditory signal to grab attention. When the vehicle first detects a pedestrian and switches to acknowledgment mode, it emits a soft chime sound. The chime is meant to serve a similar role as a driver perhaps flashing their headlights or a short horn tap to signal presence—but in a friendly manner. We were cautious not to create noise pollution or startle pedestrians; thus, the sound is subtle and used sparingly (only upon initial detection if the pedestrian has not noticed the vehicle). In later user feedback sessions, we planned to assess whether this sound was useful or could be omitted.

Vehicle Motion Cues: The vehicle’s driving behavior was programmed to reinforce the interface messages. In the simulation, when approaching a pedestrian, the AV begins to decelerate earlier and more smoothly than normal, to clearly telegraph a yielding intent through its motion (consistent with implicit communication strategies \cite{Dey2017}). The AV comes to a full stop a few meters before the crosswalk line, rather than creeping into the crosswalk, to give pedestrians space and confidence. Once the pedestrian crosses and clears the path, the AV would remain stopped for a brief buffer interval before accelerating again, and the display would indicate “Crossing complete” or revert to idle. These motion patterns are part of the interaction design, ensuring that the external HMI’s explicit signals align with what the vehicle is actually doing—any mismatch could be detrimental to trust (e.g., saying “safe to cross” while still moving forward would be unacceptable).

We implemented the above features in Carla by creating a virtual city street scene. The autonomous vehicle is represented as a sedan model equipped with a textured surface for the LED display. Pedestrian avatars can be spawned to simulate people waiting or crossing. We scripted scenarios where a pedestrian approaches a crosswalk as the AV is coming down the road: the AV’s sensors (simulated by trigger zones or raycasts in Carla) detect the pedestrian at a certain distance, then the AV’s behavior controller switches from cruise to yield mode. The external display texture animates from neutral to the eye graphic and text, and a 3D sound for the chime is played. The AV slows and stops, and the pedestrian avatar crosses. We also created variations: one scenario with the full interface active, and one where the AV is identical but with no external signals (relying only on its braking). This was to allow a comparative element in evaluation if needed.

---

This version integrates the content from your image, making the description of the system flow more naturally. Let me know if you'd like further modifications or additions!