A deep reinforcement learning (DRL) workshop for infrastructure maintenance planning was held at TUM GNI on 12th May 2023. Organized within the scope of the TUM GNI funded project INFRA.RELEARN by Prof. Daniel Straub and Prof. Kostas Papakonstantinou, the event brought together 23 leading experts to discuss the potential of DRL in optimizing the operation and maintenance of engineering systems.
Expert talks provided an overview of and insights into research and industry perspectives leading to engaging discussions on the capabilities and limitations of DRL methods, considering industry requirements. PhD students showcased decentralized DRL approaches, robust solutions, parallelized training techniques, and large-scale system implementation under constraints, which fostered debate on the most pressing research questions. Issues such as the scaling of DRL solutions for complex systems, the role of model uncertainty on DLR-based decisions, the implications of policy interpretability, and operational safety challenges were explored, and discussions touched upon practical challenges, including human intervention, feedback, regulations, and data quality. The workshop ended successfully with the agreement to establish common benchmarks for testing and comparing algorithmic solutions.
A more detailed summary of the event can be found here: https://infra-relearn.github.io