Noisy MOO

Noisy Multi-Objective Optimization

Our work on Automated Driving Systems (ADS) led the ERATO MMSD group to perform various research in the area of Search-Based Testing (SBT) (e.g. Frenetic) . However, when applying our techniques to the Autonomoose ADS, we faced a new research challenge: non-determinism.

Existing techniques to counteract such noisy execution behaviour typically require estimating the noise or development of noise models. Our work aims to discover and develop new approaches to overcome this challenge. In the process, we developed the k-nearest neighbours averaging (kNN-Avg) method, which uses an SBT algorithm’s previous execution history to produce more robust techniques.

In the process, our ongoing research shows promising results and resulted in multiple publications.

Related Publications

(Klikovits & Arcaini, 2021)(Klikovits & Arcaini, 2021)
  1. Handling Noise in Search-Based Scenario Generation for Autonomous Driving Systems
    Stefan Klikovits, and Paolo Arcaini
    In 26th IEEE Pacific Rim International Symposium on Dependable Computing (PRDC 2021), 2021
  2. KNN-Averaging for Noisy Multi-Objective Optimisation
    Stefan Klikovits, and Paolo Arcaini
    In Proc. 14th Intl. Conf. on the Quality of Information and Communications Technology (QUATIC), 2021