Insurance
Home›Insurance›Industry & Deals›Near-miss driving data improves training for automated…
Near-miss driving data improves training for automated vehicle safety
A University of Michigan team reported a 90% improvement in vehicle safety performance by focusing simulation training on rare, safety-critical near-miss scenarios.
Researchers at the University of Michigan developed a training approach for self-driving vehicle algorithms that aims to cut both the time and cost needed to prepare automated vehicles for real-world deployment, according to a study covered by Insurance Journal.
Instead of relying only on large volumes of routine driving data, the framework emphasizes “near-miss” scenarios, which are rare but safety-critical events that stress a vehicle’s decision-making in ways that typical data sets may miss.
The researchers tested the method by incorporating both safety-critical and near-miss scenarios in simulation and reported a 90% improvement in the vehicle’s safety performance.
The work builds on U-M testing and validation of connected and automated vehicles, including prior use of AI to reduce testing miles required by 99.9%, and it was funded in part by the National Science Foundation and the Center for Connected and Automated Transportation at U-M, with publication in Nature Communications earlier this year.