Why did the self-driving car crash? New AI tool may finally have the answer

Researchers at King’s College London have created an AI tool designed to uncover the exact causes of self-driving car crashes by analyzing the chain of events leading up to accidents. Unlike traditional statistical methods, which estimate crash probabilities, the new system uses 'actual causality' to pinpoint specific contributing factors, even if they occurred minutes or miles before the collision. The technology, presented at the 2026 IEEE International Conference on Robotics and Automation, employs a 'responsibility-guided' search algorithm to efficiently identify key causes without requiring excessive computational power. This approach is particularly important for autonomous vehicles, which process vast amounts of data—such as traffic, pedestrians, road signs, and weather—to make decisions in real time. The team behind the research believes the method could enhance safety by helping engineers design more reliable systems and build public trust in AI-driven technologies. While initially tested on self-driving cars, the tool may also apply to other AI-powered systems, including autonomous robots in hospitals or care homes. The work builds on earlier research by the same team, which focused on rare failure scenarios, but now provides clearer explanations for why those failures occur. The researchers emphasize that understanding accident causes is just as critical as preventing them as autonomous vehicles become more widespread. Self-driving cars are already operating in cities like London and San Francisco, but past incidents have raised concerns about their decision-making processes. This new AI tool could help address those concerns by offering detailed insights into crash causes.