Autonomous vehicles still lack clear way to communicate with pedestrians

Autonomous vehicles (AVs) are already operating in pilot programs across multiple countries, but standards for how these vehicles signal intent to pedestrians remain fragmented. A new study suggests that without stronger alignment between research, regulation, and industry practice, this fragmentation could persist well into real-world deployment.

Titled Rethinking External Communication of Autonomous Vehicles: Is the Field Converging, Diverging, or Stalling?, the study, submitted on arXiv, analyzes hundreds of studies and finds persistent fragmentation between research concepts and real-world deployment.

A decade of research, but no clear consensus

The study tracks the rapid growth of research into external communication for autonomous vehicles following early demonstrations of self-driving technology in the mid-2010s. Early studies focused on a central question: should autonomous vehicles explicitly communicate with pedestrians at all, or should they rely solely on vehicle movement and existing traffic rules to signal intent.

Over time, research expanded to explore a wide range of eHMI concepts, including light strips, displays, projections on the road surface, auditory signals, symbols, text messages, and even anthropomorphic cues designed to mimic human interaction. These concepts were tested in virtual environments, controlled laboratory settings, simulators, and limited real-world trials.

According to the authors, this expansion created a rich but increasingly fragmented research landscape. While hundreds of studies investigated variations of similar ideas, they often used different experimental designs, metrics, and contexts, making comparison difficult. Many studies focused on short-term pedestrian comprehension or perceived safety rather than long-term behavior, scalability, or regulatory feasibility.

The review finds that convergence has occurred around one principle: safety communication should be implicit rather than instructional. Research increasingly agrees that autonomous vehicles should not tell pedestrians what to do, such as instructing them to cross, but should instead communicate vehicle status and intent in a way that allows pedestrians to make their own decisions. This shift reflects growing awareness of legal liability and ethical responsibility in mixed-traffic environments.

Light-based signals have emerged as the most widely supported modality. Studies consistently find that simple visual cues integrated into vehicle lighting systems are easier for pedestrians to interpret, less distracting, and more compatible with existing traffic norms. These signals are also more likely to meet regulatory requirements, which severely limit the use of text, projections, or non-standard displays on vehicles.

Where research and reality diverge

Despite partial convergence on basic principles, the study identifies significant divergence between academic research and real-world deployment. Many experimental eHMI concepts explored in the literature have little chance of being implemented at scale due to regulatory, technical, or practical constraints.

Projections onto the road surface, for example, are frequently studied but face challenges related to visibility, weather conditions, infrastructure variability, and legal approval. Text-based messages, while effective in controlled experiments, raise concerns about language barriers, distraction, and liability. Anthropomorphic designs, such as animated eyes or gestures, attract academic interest but are largely absent from industry deployments due to concerns about misinterpretation and overtrust.

The authors highlight that industry implementations are already narrowing the design space. Vehicle manufacturers and technology companies tend to favor minimal, standardized solutions that align with existing lighting regulations and can be deployed globally. This filtering process means that much of the exploratory research conducted in academic settings may never translate into real-world systems.

The study also points to divergence in how success is measured. Academic studies often prioritize subjective measures such as perceived safety, trust, or likability, while industry and regulators focus on objective outcomes such as collision reduction, predictability, and legal clarity. This misalignment complicates efforts to translate research findings into policy or standards.

Another source of divergence lies in cultural and contextual assumptions. Many studies are conducted in controlled environments with participants who are aware they are interacting with autonomous vehicles. Real-world pedestrians, by contrast, may not know whether a vehicle is automated, partially automated, or human-driven. The authors argue that this uncertainty undermines the effectiveness of many eHMI concepts that assume high levels of user awareness.

Signs of stalling and the path forward

While the number of publications has grown steadily, many recent studies revisit similar questions using similar methods without resolving long-standing debates. The authors note repeated comparisons of the same signal types, incremental design variations, and limited attention to longitudinal effects or system-level integration.

The review suggests that the field risks diminishing returns unless research priorities shift. Instead of continuing to explore novel interface concepts in isolation, the authors argue for greater emphasis on consolidation, replication, and ecological validity. This includes testing eHMIs in more realistic traffic scenarios, examining interactions with diverse populations, and studying how communication systems perform over time rather than in one-off encounters.

Regulation emerges as a central factor shaping the future of eHMIs. The study highlights how international vehicle standards and traffic laws already constrain what can be displayed on vehicles. Rather than viewing regulation as a barrier, the authors argue that it should be treated as a design input. Research that ignores regulatory feasibility is unlikely to influence real-world outcomes.

The authors also call for clearer differentiation between universal and context-specific communication. Some signals, such as indicating whether a vehicle is yielding, may need to be standardized globally. Others could be adapted to local traffic norms, environments, or user groups. Without this distinction, efforts to standardize may either oversimplify complex interactions or fail to gain adoption.

Importantly, the study emphasizes that external communication cannot compensate for poor vehicle behavior. Predictable, rule-abiding motion remains the most powerful signal an autonomous vehicle can provide. eHMIs should be seen as complementary tools that enhance clarity in ambiguous situations, not as substitutes for safe driving behavior.

Implications for safety, policy, and public trust

Pedestrians rely heavily on subtle cues to assess risk, and uncertainty around vehicle intent can lead to hesitation, unsafe crossings, or overtrust. Inconsistent or overly expressive eHMIs risk creating new forms of confusion rather than reducing it.

From a policy perspective, the study underscores the need for closer collaboration between researchers, regulators, and industry. Without shared frameworks and comparable metrics, the field may continue to generate knowledge that is difficult to apply. The authors suggest that future progress will depend on fewer but more coordinated research efforts aligned with deployment realities.

The review also highlights a broader challenge facing autonomous systems: communication is not just a technical problem but a social one. Pedestrian behavior is shaped by culture, experience, and context, and no single interface is likely to work universally. This complexity reinforces the need for cautious, evidence-based deployment rather than rapid experimentation on public roads.