Research suggests AV pedestrian detection systems are biased

As the autonomous vehicle (AV) timeline is reconsidered and redrawn following years of what some commentators label misguided hype, an increased emphasis on safety has emerged as a key focus point. But is there an inherent threat of AV detection software bias that puts some pedestrians at greater risk based on characteristics like race and age?

It was considering this question that prompted Dr Jie Zhang, Lecturer in Computer Science at King’s College London, to investigate the subject further. “The ethics and fairness of AV systems are critical—these are things that will affect peoples’ daily lives,” she emphasises to Automotive World. However, upon searching for available information on the issue, she found there was practically none in the public domain.

Taking matters into her own hands, Zhang led a team of colleagues from Peking University and University College London to assess eight AI-driven pedestrian detection systems currently used in AV development.