Researchers have developed a new AI model called EG-ARSA to help improve road safety in low-income countries. The model uses expert knowledge from field audits and distills it into a compact vision-language model that can identify potential hazards on roads. This approach is more cost-effective and scalable than traditional methods, which rely on expensive large-scale inspections or incomplete crash records. The team also created an open dataset called BD-ARSA containing 21,947 image-audit records from Bangladesh. Experimental results show that the EG-ARSA model outperforms other AI models in identifying potential hazards. --- Why it matters: This matters because it provides a more effective and scalable solution for proactive road safety auditing in resource-constrained environments, which can help reduce traffic-related injuries and deaths. Source: https://arxiv.org/abs/2608.23563