Victor-Alexandru Darvariu

Proactive coordination in dynamic combinatorial optimization is achieved by grounding agent plans in the environment, enabling agents to outperform centralized oracles on unseen, larger instances.

Graph reinforcement learning transforms corner kick strategies from mere imitation of past plays to dynamic, reward-driven tactical innovation.

Autonomous underwater gliders can now navigate more efficiently and reliably for months at a time thanks to a new online planning system that leverages physics-informed simulation and Monte Carlo Tree Search.