Towards Learning-guided Search for Coordination of Multi-agent Transportation at Scale

A key challenge for coordination of multi-agent systems is the combinatorial nature of the joint decision space. Despite the success and popularity of model-free learning methods in various areas of autonomy, they cannot adequately tackle large multi-agent coordination problems. My research aims to ground policy learning in well-suited search-based methods to improve solution quality and reduce planning time overhead in multi-agent transportation problems, often involving task allocation, routing, path finding, and motion planning. The implications of this work has numerous applications across fields such as robotics, logistics, and transportation, paving the way for broader deployment of learning-based methods in large-scale autonomous systems and for combinatorial decision problems in general.