University of Lincoln School of Computer Science CMP2020M Artificial Intelligence
Classic AI Planning
Open the Planning Web Service at http://lcas.lincoln.ac.uk/fast-downward. For all the following tasks discuss your main findings and observations with your neighbours and the demonstrators. A planning problem is described by a ¡°domain¡± specification (defining all the actions and their preconditions and effects) and a ¡°problem¡± (describing initial and goal states). These are given in the language PDDL, a language inspired by logic languages like LISP and Prolog.
Q1.Read the introduction on PDDL provided on blackboard.
Q2.Familiarise yourself with the interface and load the domain and problem specifications for the ¡°blocksworld¡± from the pull-down menus (It¡¯s called ¡°blocks¡±).
a. Look at the predefined problem and
draw (similar to the figure on the right) the initial state and the goal state defined in the problem file.
b. Find a plan yourself (not using the planner) that changes the initial state into the goal state. How long is your best plan? Is it the only possible solution to achieve the goal state? What strategy do you use to find a plan?
c. Let the web service solve the given blocks problem. Compare to your own plan. Is it the same?
d. Now change the given blocks problem in the web service to represent the initial state shown in the top figure, and make the goal state corresponding to the figure on the right. Let the web service find a plan and check it makes sense.
Q3.Look at the ¡°robot example on the webservice and understand it. Add another roomc, and ask the robot to put two of the balls in that room as a goal.
Q4.(optional challenge) Look at this classic puzzle:
=>
Define the problem and domain in PDDL; hint: define actions as movements of
the empty space, and enumerate the four places.