Self-learning systems for robust optimal performance


In 2023, key activities are centered around showcases and research lines, notably Vehicle Energy Management (SC1) and Space-based optical communication terminal (SC2). SC1 focuses on developing battery-conscious energy management to increase battery life by 10%, while SC2 works on terminal-level models for awareness and optimization in varying atmospheric conditions. The research lines involve the development of probabilistic awareness (RL1), cross-domain self-learning methodology for system optimization (RL2), and a taxonomic study on system adaptation loops (RL3). RL2 showcases the use of self-learning risk assessment in SC1 and SC2, optimizing solutions and satellite controller parameters. The taxonomic study in RL3 identifies elements influencing system adaptation, showcasing how optimization methodologies adapt systems under risk factors and time horizons, such as adjusting constraints in SC1 missions based on past battery depletion to extend battery life.


The AutoADAPT awareness concept was developed and demonstrated for the electric truck showcase, with a probabilistic reasoning model that exceeded previous results in detail and accuracy. The satellite communication showcase includes the first version of the optimization solution (TRL 2), defining the correct configuration parameters for the feedback controller. The automotive showcase features a novel self-optimization solution based on a genetic algorithm, considering both short and long time horizons. For these time horizons, an Adaptation concept with two corresponding loops is developed. A scalable computation platform is realized, which support fast computations.


  • Rene Corbeij, Senior Project Manager, e-mail: