The Blind Spot in Self-Driving Cars: High-Visibility Clothing

High-visibility clothing, traditionally used for pedestrian and cyclist safety, may not be as effective with autonomous vehicles (AVs). AVs rely on sensors and algorithms, unlike human visual perception, which can lead to detection issues.

Recent studies and incidents, such as the failure of AEB systems to detect pedestrians in reflective clothing, highlight these concerns. Additionally, real-world incidents have shown AVs failing to avoid pedestrians in several cases.

As AV technology evolves, pedestrians should remain cautious and not solely rely on high-visibility clothing for safety.

If you've driven by a construction site, chances are you've seen the workers wearing brightly-colored uniforms - also known as high-visibility clothing.

High-visibility clothing has long been a very important aspect of pedestrian and cyclist safety, designed to get the attention of human drivers and reduce accidents. According to the American National Standards Institute, the approved clothing typically includes colors such as fluorescent lime, orange, and red.

However, new research and recent incidents suggest that these bright and reflective clothes may not offer the same level of protection against AI-driven cars.

The Evolution of Road Safety Measures

For decades, safety for pedestrians and cyclists on the road have relied on human eyes. Fluorescent materials and reflective strips are highly effective in capturing the attention of drivers, especially when it's dark out.

Yet, autonomous vehicles (AVs) don't see the world as we do. Instead, they use sensors, including cameras, LiDAR (Light Detection and Ranging), radar, and infrared systems, to detect and interpret their surroundings.

This fundamental difference in perception raises a critical question: Does high-visibility clothing still serve its intended purpose with self-driving cars?

While human drivers instinctively respond to bright colors and reflective gear, AVs rely on algorithms that analyze shapes, movement patterns, and contrasts. This means that the effectiveness of safety gear may be compromised, especially where there are multiple objects competing for an AI’s attention.

Recent Findings on High-Visibility Clothing and AVs

A recent study by the Insurance Institute for Highway Safety (IIHS) has brought this issue into focus. Researchers tested the pedestrian detection systems of several 2023 vehicle models equipped with Automatic Emergency Braking (AEB) systems. The tests involved mannequins dressed in various outfits, including high-visibility jackets with reflective strips.

The findings were concerning. The AEB systems in both the Honda CR-V and Mazda CX-5 failed to detect pedestrians wearing reflective clothing, particularly under low-light conditions.

Only the Subaru Forester demonstrated a higher detection rate, successfully avoiding collisions in most scenarios except those with extremely low visibility.

Another study conducted by the University of California, Berkeley, found that AI systems in self-driving cars are more reliable at detecting pedestrians with distinct movement patterns rather than those relying solely on visibility.

The study suggested that AVs struggle in environments with visual noise—such as construction zones, busy intersections, and foggy conditions—where bright clothing may not be enough to distinguish a pedestrian from surrounding elements.

Real-World Incidents With Detection Failures

In October 2023, a pedestrian in San Francisco was struck by a human-driven vehicle and subsequently run over by a Cruise autonomous vehicle operating without a driver. The AV reportedly attempted to avoid the pedestrian by swerving but ultimately failed to prevent the collision, dragging the individual for approximately 20 feet before coming to a stop.

Similarly, the tragic death of Elaine Herzberg in 2018, who was struck by a self-driving Uber vehicle in Arizona, highlighted shortcomings in AV pedestrian detection systems. Investigations revealed that the vehicle's sensors detected Herzberg six seconds before the collision but misclassified her as an unknown object, then as a vehicle, and finally as a bicycle, leading to a delayed response.

Another concerning case occurred in 2022 when a Waymo autonomous vehicle failed to recognize a construction worker directing traffic with a reflective vest and a stop sign.

The vehicle hesitated and ultimately proceeded through the intersection, forcing the worker to jump out of the way.

Investigations into Autonomous Driving Systems

The National Highway Traffic Safety Administration (NHTSA) initiated an investigation into Tesla's "Full Self-Driving" system following reports of crashes under low-visibility conditions, including a fatal incident involving a pedestrian.

The probe covers approximately 2.4 million Tesla vehicles produced between 2016 and 2024 and aims to assess the system's response to reduced visibility scenarios.

Tesla's approach to autonomous driving relies heavily on computer vision and end-to-end machine learning, omitting additional safety layers like radar and LiDAR used by other manufacturers. While this strategy allows for extensive data collection and potential cost reductions, it raises concerns about safety and regulatory approval due to the technology's complexity and potential for error.

In response, companies like Waymo and Cruise have been incorporating additional sensors and refining their AI models to better recognize pedestrians in different conditions. However, there is still progress to be made, and critics argue that more stringent testing protocols are needed before AVs are allowed out on public roads.

Possible Solutions to Improve Pedestrian Safety

As AV technology advances, researchers and engineers are exploring ways to enhance pedestrian detection capabilities. Some potential solutions include:

• Improved Sensor Fusion: Integrating multiple sensor types, such as LiDAR, thermal imaging, and radar, can improve pedestrian detection in low-visibility conditions.

• AI Training on Diverse Datasets: Developers are working to ensure that AV systems are trained on diverse datasets featuring pedestrians in a variety of clothing, lighting conditions, and movement patterns.

• Smart High-Visibility Clothing: Some researchers are developing clothing embedded with digital markers or sensors that communicate directly with AVs, ensuring detection regardless of lighting conditions.

• Standardized Safety Regulations: Governments and regulatory bodies are pushing for higher safety standards and requiring manufacturers to prove their systems work effectively across different real-world scenarios.

Implications for Pedestrian and Cyclist Safety

As we've likely all experienced with our phones or computers, technology is not always 100% fool-proof. The same goes for the technology in self-driving cars, making it concerning to be on the road with them.

Although technology is advancing everyday, it's tough to see past the issues with high-visibility clothing detection, when these uniforms are specifically designed to be seen in almost any condition.

There's still a lot of progress to be made on the manufacturing side as well as the regulator's side.

In the meantime, pedestrians and cyclists should stay aware of their surroundings and not solely rely on high-visibility clothing for safety. Other precautionary measures, such as making eye contact with human drivers, using pedestrian crossings with traffic signals, and staying aware of AVs’ limitations, can help reduce risks.