Chinese automakers are developing vehicles that sense, decide, and respond before their occupants do. But a function that can be built is not necessarily one that is safe or comfortable for people. From exterior-light colour and luminance to the healthy cockpit, Fudan University's Lin Yandan showed how human response can become part of automotive lighting validation.

By Sang Min Han _ han@autoelectronics.co.kr
한글로보기






A car moves down a dark road with a blue indicator light on its exterior. In China, it became known as the "little blue light" and spread rapidly as an external indication that a Level 2 driver-assistance system was active. To many consumers, its illumination came to suggest the beginning of autonomous driving. China's policy direction, however, drew a clear line: Level 2 remains driver assistance, not automated driving in which the vehicle assumes the driving task. In July 2026, China moved to stop its use on newly certified passenger vehicles. The color fell outside what the country's exterior-lighting standard, GB 4785, permits, and the light had drawn complaints over nighttime glare and interference with other drivers' judgment.
Removing the blue light does not remove the need for a vehicle to communicate its automated state and intended actions. Matrix LEDs, pixel lighting, and road projection may provide new forms of external HMI. But a different display technology does not change the questions that must be answered. Does a person notice the signal faster? Do they understand what it means? Does making it brighter or more elaborate actually make anyone safer?
Lin Yandan's research was not about whether one particular indicator light should be permitted. It measured those questions directly, through human response. In her experiment, blue attracted attention more quickly than white. But once brightness crossed roughly 102.5 cd/m², identification stopped improving in any meaningful way, while cognitive load and glare discomfort continued to increase. The technology could make the light brighter. People did not get more information out of it.




Blue signal lights showed 24% higher contrast and a 14% higher vehicle-identification rating than white lights. However, combining this advantage with higher brightness introduced another problem.



Blue Was Noticed Faster — But Only Up to a Point

A self-driving car has no eye contact or hand gesture to offer. Pedestrians and other drivers still need to know whether it is about to move off, stopping, yielding, or currently driving itself. That is the point where a vehicle's exterior lighting moves past decoration or basic illumination and becomes an eHMI — a channel for communicating the vehicle's state and intent.
Lin presented a study comparing blue and white indicator lights signaling autonomous status. Participants took less time to first notice the blue light than the white one. The reported contrast effect was 24%, average fixation duration was longer, and vehicle-identification ratings were 14% higher for blue than for white. The blue light attracted attention more quickly. But catching someone's eye faster does not by itself mean the risk of an accident is lower.
Her team ran a nighttime observation study with 26 participants and an actual vehicle fitted with the indicator light. Brightness was tested across four levels from 101.5 to 103.75 cd/m², combined with three mounting positions and three viewing distances. Both brightness and mounting position produced statistically significant main effects on cognitive load (p<.001 for each). A light mounted along the center line produced noticeably lower cognitive load than one along the windshield line or across the grille. Past roughly 102.5 cd/m², identification did not improve further, while cognitive load and glare discomfort continued to rise.
One line on her slide summed it up: "Rewriting the simple logic that more visible means safer."
"You cannot judge safety by color alone," she said. "Lighting safety depends on color, brightness, mounting position, viewing distance, and the real traffic environment together."
102.5 cd/m² is not a universal safety threshold for every autonomous-status light on the road. It is a figure drawn from one vehicle, one nighttime setup, and 26 participants. Change the age group, visual characteristics, weather, or road environment, and the result may shift. Turning it into a standard or a product specification would require testing across a much wider range of conditions. What the figure demonstrates is not a fixed answer — it is that the maximum a technology can deliver and the level appropriate for people are not necessarily the same.
This study was one piece of a larger day of presentations. Desay SV described an AI agent that reads a user's intent and links vehicle functions together on its own. XPENG described seats that sense a passenger's posture and physical state and respond before being asked. Dongfeng laid out an intelligent chassis that anticipates the limits of the road and the vehicle in advance. Across the sessions, one direction repeated: the vehicle senses, decides, and acts before the person does.
The opening slide of Lin's own talk read almost like a footnote to that direction. Automotive lighting, she said, has entered a system-level transition. With electrification, intelligence, connectivity, and experience converging at once, a light's basic function is no longer limited to being seen and letting people see. She listed roles spanning environmental sensing, information exchange, safety warnings, emotional expression, and health monitoring — and the competencies needed to support them extended beyond optical design into materials, measurement and validation, standards and regulation, and human factors. Her indicator-light experiment was one example within a single branch of that framework — measurement and validation.




A nighttime real-vehicle observation study involving 26 participants found that identification performance stopped improving above 102.5 cd/m², while cognitive load and glare continued to increase.



When the Product Changes, So Must the Way It Is Measured

Automotive lighting can no longer be evaluated as a single-function product in a fixed, finished form. Lamps now stretch more than two meters across the front of a vehicle, with multiple functional zones and dense LED arrays built into one illuminated surface. Existing methods still measure luminous intensity and beam distribution at a limited number of regulatory test points. Lin summed up the resulting gap this way: "New challenges are forcing the whole system to keep revising itself." Regulation does not lead the change. Products are changing first, leaving measurement methods and standards to follow.
On long illuminated surfaces, there is no single point that can serve as the light's center. Hangzhou-based EVERFINE Corporation proposed using machine vision to locate the spatial coordinates of individual light-emitting points, combining image-based measurement with coordinate transformation to identify positioning errors in both horizontal and vertical directions. HUDs present a similar difficulty: every vehicle's curved glass and display area differs, so fixed test rigs struggle to keep pace. Shanghai-based Fuzhan Intelligence built a flexible inspection architecture instead, covering production, aftermarket service, and export delivery together. Combining a multi-axis platform, image algorithms, and data traceability, this approach is shifting inspection from single-point parameter checks toward quality control across the entire product lifecycle.
Sunlight focusing presents another reliability challenge that initial optical-performance testing may not capture. A lens can concentrate sunlight from a specific angle into a small, high-energy spot, discoloring or deforming the material inside — and in severe cases, disabling the lamp's function entirely. Lin described a three-stage test procedure: pre-scanning with a light source that approximates natural sunlight, a full sweep across different elevation and azimuth angles on the lens module or complete lamp assembly, and reproducing high-temperature conditions to evaluate resistance under extreme scenarios. She proposed that sunlight focusing be written into the full-lifecycle safety validation standard for lamps.
The talk did not stop at measurement. She pointed to a silicone collimator for pixel headlamps and a polyurethane hot-melt adhesive as examples of how a smart lamp's competitiveness depends on cost, manufacturing efficiency, and long-term reliability, not optical performance alone. Building smarter lighting at scale, she argued, requires rethinking the materials that make up and seal the lamp itself. The scope of measurement cannot stop at the point of shipment either. The more sensors, controllers, and software a smart lamp carries, the more there is to evaluate — heat, materials, electronics and algorithms, manufacturing variance, and years of use in the field.




DLP-based interaction projects pedestrian-crossing guidance directly onto the road. However, uneven road surfaces and the risk of misinterpreting its meaning remain unresolved challenges.



Lighting Carries Meaning, Not Just Signals

Lin argued that future vehicle lighting will not function as an isolated illumination or signaling system. It has to work together with sensors, domain controllers, cameras, the road environment, and cloud data. Pixel lighting, discussed above, is one example: adaptive driving beam (ADB) applies that pixel-level precision to real-time environmental sensing, shading only an oncoming vehicle rather than dimming the beam as a whole. An autonomous-status indicator light communicates the driving mode in a way that satisfies both regulatory requirements and outside visibility, while an eHMI interprets intent in terms a human can follow, organizing information for pedestrians and other road users.
"The vehicle lighting system of the future needs to move beyond 'lighting the road,' toward 'organizing information' and 'interpreting intent,'" she said. "Clear, stable, understandable light signals are what reduce the information gap between everyone sharing the road."
DLP projection lighting follows the same logic. A vehicle can project a lane-change alert, a blind-spot warning, or a crosswalk cue directly onto the road surface. But road irregularities and changes in the vehicle's posture can blur or distort the projection, a misread signal can turn into a safety hazard, and how far regulation should extend here remains unsettled.
"Projected interaction cannot remain an idealized rendering," she said. "It has to be validated on real vehicles and real roads, with a standardized evaluation framework built around it."




The health cockpit aims to quantify and validate sensory concepts such as “health” and “comfort.” The CCRT framework translates actual vehicle-owner feedback into measurable product-definition metrics.



The Person Becomes Lighting's Next Spec

In the closing part of her talk, Lin's scope extended to areas lighting does not directly touch. Autonomous driving, she argued, does not remove the human role — it shifts a person from driving one vehicle directly to monitoring several vehicles and intervening when needed.
In remote operation of driverless vehicles, the operator shifts from a direct driving role to that of a remote supervisor, and eye movement, brainwave, and near-infrared signals together become a composite indicator for tracking cognitive fatigue and operational risk.

Visibility alone was not enough. The next question was how light affects human physiology and cognition. Southwest Jiaotong University's research showed that the photobiological effects mediated by intrinsically photosensitive retinal ganglion cells (ipRGCs) influence alertness, cognitive performance, and emotional state — evidence that lighting reaches past visual signaling into a person's physiological rhythm. Shanghai Ocean University's lightweight relationship-aware network, MEXNet, addressed the issue from the algorithmic side, building lighting-related human-factors analysis directly into its algorithm design and demonstrating how much the actual light environment shapes the performance of driver monitoring systems (DMS). A detection algorithm that performs accurately in the lab may perform poorly under backlighting, at night, in tunnels, or under uneven illumination — the same issue raised earlier by her indicator-light study.
The healthy-cockpit segment introduced a framework called CCRT, presented as an "engineering translator" for user experience. It combines owner feedback, subjective assessments, physiological indicators, and objective measurements with full-scenario vehicle testing and multi-source analysis. The output is a set of product-definition metrics and identified user pain points, which converged on four areas: space, dynamic ride comfort, motion sickness, and seat pressure. Interior lighting fell within the same frame — interaction that adjusts dynamically to driving scenario and passenger needs, and the comfort and mood regulation that brightness, color temperature, and dynamic lighting patterns provide. Lighting, too, became something to quantify under "health" and "comfort."
"A healthy cockpit has to bring together environment, temperature and humidity, vibration, seat support, visual load, and multiple streams of physiological data," she said. "We need evaluation methods that let sensory concepts such as 'healthy,' 'comfortable,' and 'premium' be quantified, validated, and engineered."



Back to 102.5 cd/m²

Lin treated automotive lighting as a system, but people remained the measure of its performance. The significance of 102.5 cd/m² lies there. The light could be made brighter, but beyond that point participants did not identify it more effectively; cognitive load and glare discomfort continued to rise.
Regulation can remove the little blue light. It cannot remove the task the light was meant to perform. The next generation of eHMI must still show whether people notice its signals quickly, understand them correctly, and receive them without unnecessary burden. The number is not an answer. It shows that the validation is unfinished. As vehicles increasingly sense, decide, and act before people do, people must remain the final measure.

AEM(오토모티브일렉트로닉스매거진)