Self-driving cars may not pick up pedestrians depending on what they’re wearing, experts warn
Self-Driving Cars Could Miss Pedestrians Based on Clothing, Experts Warn
Self driving cars may not pick - As London prepares to expand its autonomous vehicle trials, concerns have emerged about the ability of self-driving cars to reliably identify pedestrians. Experts, including AI specialists, are raising alarms about how clothing choices might influence detection accuracy, potentially leaving some individuals overlooked by the technology. This issue comes to light as companies like Waymo seek to introduce commercial self-driving taxi services by the end of 2026, sparking debates over safety and system reliability in diverse urban settings.
London Assembly Examines Challenges in Automated Vehicle Safety
During a recent London Assembly session, officials discussed the growing challenges of ensuring automated vehicles operate safely in mixed traffic environments. A key concern was how current systems might misidentify pedestrians, particularly those wearing certain types of clothing that could blend into surroundings or appear similar to stationary objects. This problem is exacerbated in cities with high pedestrian activity, where subtle visual cues can determine whether an autonomous car reacts appropriately.
“We’ve seen experimental evidence from Canadian collaborators showing that pedestrians’ clothing can affect whether self-driving cars recognize them,” said Professor Siddartha Khastgir, head of Safe Autonomy at the University of Warwick’s Warwick Manufacturers Group. “For example, winter attire might be mistaken for static elements in the environment, reducing the system’s responsiveness.”
Prof Khastgir emphasized the importance of inclusive training data to address these biases. “Automated vehicles must be trained to identify people of all body types and attire,” he added, highlighting that systems should not rely solely on visual patterns like reflective gear or bright colors. This call for diversity in training data underscores the need for algorithms to adapt to real-world variations, ensuring safer navigation in London’s dynamic streets.
Research Reveals Disparities in Detection Accuracy
Studies from King’s College London further reveal how self-driving vehicles may exhibit inconsistent detection rates. Findings indicate that these systems are 20% more likely to spot adults than children, and 7.5% more likely to recognize white individuals compared to ethnic minorities. Such disparities suggest that the algorithms may favor certain visual features, like light-colored clothing or typical body shapes, over others, leading to potential safety risks for underrepresented groups.
Prof Khastgir explained that the training data used by AI models often lacks diversity, creating blind spots. “If the data doesn’t reflect the full range of pedestrian appearances, the system might prioritize familiar patterns at the expense of accuracy,” he noted. This could result in pedestrians wearing less noticeable attire or non-standard clothing being missed, increasing the likelihood of collisions or near-misses in busy areas.
Elly Baker Challenges Waymo’s Safety Claims
Labour’s Transport Spokesperson, Elly Baker AM, questioned the safety measures proposed by Waymo during the assembly meeting. While the company highlighted scenarios like detecting movement in the cabin, Baker pointed out that these examples don’t fully address subtler risks, such as passengers being assaulted without visible signs of distress. “The example given doesn’t replace what a human in the front would be able to notice in the back,” she argued, stressing the need for more robust safety protocols.
“We need to ensure automated systems are equipped to handle more than just obvious signs of movement,” Baker added. “This includes identifying uncomfortable interactions that may not involve physical activity or seatbelt changes.”
Baker’s critique highlights a broader debate about whether self-driving cars can fully replace human oversight. While the technology offers promise, experts agree that refining detection algorithms and incorporating diverse datasets are critical steps to address these concerns and improve public trust in autonomous vehicles.