The overarching goal of MENTOR is to combine Multi-Agent Reinforcement Learning (MARL) with control theoretic approaches for the development of new more efficient strategies for the control of autonomous multi-agent systems and explore their application to the problem of designing cooperative agents able to perform challenging search-and-rescue operations in uncertain environments. 

The specific objectives of the research program are:

• to carry out a comparative analysis of existing control and MARL strategies for complex multi-agent systems in the literature and their application to SAR scenarios to identify a set of strategies to benchmark those we will develop in MENTOR;
• to develop learning-based strategies to steer the collective behaviour of complex multi-agent systems combining control strategies with reinforcement learning algorithms to achieve better efficiency and shorter learning times in complex situations;
• to analyse the convergence and robustness properties of the strategies on a set of testbed applications and
• to use the new strategies developed within the project to multi-agent Search and Rescue problems in  uncertain environments such as those that occur in natural disasters' areas and validate their effectiveness through a set of simulations and proof-of-principle experimental tests.

The project is highly focused on combining control theoretic tools with machine learning. It therefore comprises two Reaserch Units at the University of Naples Federico II and at the University of Bologna with complementary expertise and know-how on control  techniques, reinforcement learning, mathematical modelling, complex systems, herding and autonomous systems. It will also exploit the strong existing research connections with the group led by Professor Michael Richardson at Macquarie University Sydney which has unique expertise on herding and distributed decision making and with the Dynamical Systems Laboratory at New York University led by Professor Maurizio Porfiri, a world renowned expert on the the modelling and experimental investigation of complex systems and animal behaviour and their rescue from natural disaster areas (e.g. after oil spills).

The project addresses two fundamental themes of the PNRR: foundational aspects of artificial intelligence with particular emphasis on its control implications, and environmental, natural and anthropic risks, with attention to novel strategies to address post-event search and rescue scenarios via the deployment of autonomous multi-agent systems. Interdisciplinarity is core to this project, with the aim to achieve a strong scientific, technological, societal and economic impact well beyond the application domain selected as a testbed scenario.