This thesis presents a novel framework for safe online trajectory planning of unmanned vehicles through partially unknown environments. The basic planning problem is formulated as a receding horizon optimization problem using mixed-integer linear programming (MILP) to incorporate kino-dynamic, obstacle avoidance and collision avoidance constraints. Agile vehicle dynamics are captured through a hybrid control architecture that combines several linear time-invariant modes with a discrete set of agile maneuvers. The latter are represented by affine transformations in the state space and can be described using a limited number of parameters. We specialize the approach to the case of a small-scale helicopter flying through an urban environment. Next, we introduce the concept of terminal feasible invariant sets in which a vehicle can remain for an indefinite period of time without colliding with obstacles or other vehicles. These sets are formulated as affine constraints on the last state of the planning horizon and as such are computed online. They guarantee feasibility of the receding horizon optimization at future time steps by providing an a priori known backup plan that is dynamically feasible and obstacle-free.