Download Poster () - Colorado State University Computer Science

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Computational electromagnetics wikipedia , lookup

Artificial intelligence wikipedia , lookup

Transcript
Fast Comprehensive Planner for Fully
Observable Nondeterministic Problems
Andres Calderon Jaramillo – Faculty Advisor: Dr. Jicheng Fu
Department of Computer Science, University of Central Oklahoma
INTRODUCTION
Research of artificial intelligence planning
aims to design planning algorithms (i.e.,
planners), which are targeted at finding plans
to take a system from an initial state to a goal
state. In this project, we propose an algorithm
that uses an existing classical planner to
efficiently find strong and strong cyclic plans for
nondeterministic problems in fully observable
domains. Additionally, we equip our planner
with heuristics that inform the search for a plan
along relevant directions. Our experiments
suggest that our algorithm significantly
outperforms state-of-the-art planners in
commonly used benchmark domains.
This work was supported in part by the National
Institute of General Medical Sciences of the
National Institutes of Health through Grant
Number 8P20GM103447.
3. Unsolved states along the weak plan are
solved in the same manner. The process is
repeated until no unsolved states remain.
Initial
State
Goal
Figure 2(a). A weak plan.
There is at least one successful path to a goal.
Constraints
Initial
State
Goal
Figure 2(b). A strong cyclic plan.
A plan may use actions that can cause cycles but
every state has a path to a goal.
In the context of artificial intelligence planning,
a domain is a structure that describes the
possible actions that can be used in finding a
plan. A planning problem for a given domain
specifies the initial state of a system and a set
of goals to achieve. A planner is an algorithm
that solves a planning problem by finding a
suitable set of actions in the domain to take the
system from the initial state to at least one
goal state.
FOND (Fully Observable Nondeterministic)
problems assume that each state in a system
can be fully observed and that the actions in
the domain may have more than one possible
outcome. Solutions can be classified in three
categories (Cimatti et al. [1]): weak plans,
strong cyclic plans, and strong plans. See
Figure 1 and Figure 2.
Initial State
C
put-down(B)
A
Goal
B
C
A
pick-up(B, A)
C
Heuristics
The following heuristics are used in addition to
those embedded in the classical planner. The
heuristics are explained in (Fu et al. [3]):
State reuse: the classical planner is stopped if
a solved state is found.
BACKGROUND
B
A strong plan search prevents the formation of
cycles in the solution by temporarily disabling
actions. Unsolvable states are handled using a
backtrack procedure.
B
A
Reused state
Initial
State
Goal
Figure 2(c). A strong plan.
The goal is achieved from any state without using
actions that cause cycles.
OUR PLANNER
Goal alternative: the search is first directed to
a solved sibling instead of the original goal.
Given a domain and a problem, our planner
attempts to find a strong solution. If one does
not exist, it proceeds to search for a strong
cyclic solution.
Solved sibling
Principles of Operation
Our basic approach is based on (Kuter et al.
[2]):
REFERENCES
1. The domain is relaxed into a deterministic
domain by splitting nondeterministic actions
into individual outcomes.
[1] A. Cimatti, M. Pistore, M. Roveri, and P. Traverso,
"Weak, strong, and strong cyclic planning via
symbolic model checking," Artif. Intell., vol. 147,
pp. 35-84, 2003.
2. A classical planner is used to solve the initial
state (find a path to the goal). The path is
then expanded into a weak plan according to
the original domain.
[2] U. Kuter, D. Nau, E. Reisner, and R. P. Goldman,
"Using classical planners to solve nondeterministic
planning problems," in 18th International
Conference on Automated Planning and
Scheduling (ICAPS), 2008.
Unsolved!
Figure 1. Example of a simple FOND plan.
The action pick-up(x, y) is nondeterministic as it can
succeed or fail (block x may fall on the table). The
action put-down(x) is deterministic.
Omitted classical plan
Initial
State
Goal
Unsolved!
[3] J. Fu, V. Ng, F. B. Bastani, and I.-L. Yen, "Simple
and fast strong cyclic planning for fully-observable
nondeterministic planning problems," in
Proceedings of the Twenty-Second international
joint conference on Artificial Intelligence - Volume
Three, Barcelona, Catalonia, Spain, 2011.