Stochastic Shortest Path

Stochastic Shortest Path (SSP) problems involve finding the optimal path through a probabilistic environment to a goal state while minimizing cumulative cost. Current research focuses on developing efficient algorithms, such as those based on value iteration, linear programming, and heuristic search, that address challenges like risk aversion (e.g., using Conditional Value-at-Risk), handling constraints, and scaling to multi-agent scenarios. These advancements are driven by the need for robust and efficient planning in high-stakes applications like autonomous navigation and supply chain optimization. The field is also actively exploring theoretical limits of learning in SSPs, particularly in offline settings and under various assumptions about the underlying environment.