New Method Unveiled for Safer, Uncertainty-Aware Controllers
Researchers have unveiled a novel method for constructing safer, uncertainty-aware controllers. The findings were presented at the 13th NASA Formal Methods Symposium in May 2021. The method, developed under the ERATO MMSD Project funded by JST, can be applied to various controller systems, including autonomous vehicles.
The method, detailed in the paper 'Automatic Transformation of Uncertainty-Unaware Controller Models into Robustified Models' by Peter Hoffmann, Rene Ester, and Dirk Matheja (presented at TACAS), enables more systematic and effortless construction of controllers that can handle uncertainty. It consists of two steps: uncertainty injection and robustification. The process generates a formula representing the limit of uncertainty the output controller can tolerate.
The research aims to help developers apply formal modeling approaches to realistic software, focusing on the core behavior of controllers. The method can be applied to various controller systems interacting with external environments in the future.
The method, presented at the 13th NASA Formal Methods Symposium, offers a systematic way to create uncertainty-aware and safe controllers. It has been successfully applied to autonomous vehicles and can be extended to other controller systems. The research, funded by JST, marks a significant step towards safer controller software.
