Given a planning domain and a planning problem, how can the robot autonomously find a plan, i.e. a sequence of actions, in order to go from the initial state to a state that satisfies the goal?
Algorithms that do this are called planners.
Three Main Types of Planners
Domain-specific planners
- Made or tuned for a specific planning domain
- Most successful real-world planning systems work this way (Mars exploration, sheet metal bending, playing bridge, etc.)
- Won’t work well (if at all) in other planning domains
Domain-independent planners ⇒ focus of most research
- In principle, works in any planning domain
- In practice, need restrictions on what kind of planning domains it can be applied
- e.g. Classical Task Planning
Configurable planners
- Domain-independent planning engine
- Input includes info about how to solve problems in some domain