Partially Observable Markov Decision Processes (POMDPs) provide a general framework for sequential decision-making under uncertainty, where an agent must act without direct access to the underlying state. We ask how such decision-making can be extended to settings where actions and trajectories must satisfy safety, resource, and operational constraints. We study constrained POMDPs, with a particular focus on online planning in continuous state and action spaces, and examine how constraints can be incorporated into Monte Carlo Tree Search methods.   Advisor: Prof.  Vadim Indelman