Needs model: Describes if the approach needs a model to be usable. This can be in the form of PDDL, PDDL-like, or other non-formal information. Usually, approaches that have this model available perform better than others as they do not have to learn a model first. However, it is not always easy to get a model. Here "x" means the approach needs this information, and "-" means the approach does not need the information.
Needs Probabilities: Describes if the approach needs to know probabilities of the non-deterministic actions to function. Like the approach before, this information is often not easy to obtain or even impossible in a dynamic scenario. Here "x" means the approach needs this information, and "-" means the approach does not need the information.
Needs Training: Describes if and how much training the approach needs before it can be used. "x", means training is required before the approach can be used, e.g., in a simulation, "-" the approach either does not need training or will learn during operation.
Learns during operation: Describes if the approach can learn during operation. "x" means that the approach will learn during operation, "-" means the approach is fixed during operation.
Performs Exploration: Describes if the approach can perform exploratory actions. This is useful if only partial information is available, and for example, new actions are tried or actions for knowledge gain are performed.
Failure resilient: Describes if the approach can handle unforeseen circumstances. For example, if the approach is able to adapt if a fault occurs during operation. "x" means the approach can deal with unforeseen circumstances, "-" means the approach will fail if an unforeseen circumstance occurs.
Guarantees: Describes if the approach gives some guarantees. 
FF-Replan: Because FF-Replan only selects actions with a non-zero percent chance of leading to a goal, FF-Replan is guaranteed to reach the goal eventually if there are no dead ends and the environment is equivalent to the provided model.
RBL: Similar to FF-Replan also RBL takes only actions that have a chance to lead to a goal. However, because RBL also updates the reliability of the actions, the environment does not have to be equivalent to the provided model. Therefore, we can guarantee that an agent with RBL will eventually reach the goal if there are no dead ends and there is a possible action sequence possible in the model that would lead to the goal. Meaning, as long as there are redundancies in the model of which not all are blocked.

Related Work

Nilson presented in his work \cite{j1994} with Teleo-Reactive programs a formalism for action sequences an agent can take to reach goals in uncertain environments. Teleo-Reactive programs are an ordered list of actions with preconditions. The first action of this list which preconditions are met, is executed indefinitely. Through clever construction of Teleo-Reactive programs, an agent can then deal with uncertainty.