Short answers, pulled from the story.
Automated planning and scheduling is a branch of artificial intelligence concerned with finding strategies or action sequences for execution by intelligent agents, autonomous robots, and unmanned vehicles. Unlike classical control and classification problems, solutions must be discovered and optimized across multidimensional space. The field is also related to decision theory.
The Classical Planning Problem is the simplest form of planning, defined by a unique known initial state, instantaneous and deterministic actions, one action permitted at a time, and a single agent. Because the starting state and all actions are fully known and predictable, plans can be expressed as straightforward sequences without branching or conditional logic.
Contingent planning applies when the environment is observable through sensors that can be faulty, making each plan step a decision tree rather than a single action. Conformant planning is more restrictive: the agent cannot make any observations at all and holds beliefs it cannot verify. Haslum and Jonsson showed conformant planning is EXPSPACE-complete.
Michael L. Littman demonstrated in 1998 that with branching actions, the planning problem becomes EXPTIME-complete. A particular case, FOND problems with goals specified in LTLf, is always EXPTIME-complete, and 2EXPTIME-complete when the goal is specified with LDLf.
The Hubble Space Telescope uses two automated planning systems: a short-term system called SPSS and a long-term planning system called Spike. These manage the telescope's complex observation scheduling.
A domain-independent planner accepts a domain model and a specific problem as inputs, allowing it to solve planning tasks across diverse areas such as block-stacking, logistics, workflow management, and robot task planning. A domain-specific planner, such as a route planner, is built to handle only one type of problem.