Thomas Sj

1
Thomas Sjöland, Per Kreuger, Emil Åström, Per
DanielssonCOL/SICS
Coordination of Scheduling and Allocation Agents
(extended abstract)
2
Complex Operations Laboratory (COL)
- focus on applications - scheduling and
planning in transportation and semi-continuous
production. - interpretation and modeling of
data with complex intrinsic properties -
forecasting transportation costs
3
Railway Application TUFF-3

• Support tool for coordination of planners in
  goods transportation
• mOzArt, distributed Oz system
• Finite Domain constraints
• Agent model
• Spaces modeled with FIPA (agent-protocol)

4
Rail Network Sweden
5
Schedule
6
Agent Model
Sales dept.
Engine rostering
Staff rostering
Plan mgr.
Track scheduling
7
Roster (one engine type)
8
Roster (three engine types)
9
CCP in coordination?

• CCP provides very natural synchronization

Reasoning with separate constraint solvers
But - same domain required
10
Role for tCCP?
Reasoning with time on metalevel - combining
simulation and scheduling Might be useful in
planning with sliding time-window
11
CLP/CCP systems from SICS

• SICStus Prolog
• global constraints
• agent platform

• (distributed) Oz /mOzArt
• Oz for networked programs mOzArt
• Finite domain constraint solvers

Can a timed approach be integrated?
12
Distributed OZ mOzArt

• Transparent distribution
• mobile objects (and code)
• http objects
• distributed unification
• distributed garbage collection
• sites connect and disconnect dynamically

13
Abstract Constraint Programming

• Translation with abstraction
• Formally motivated abstract models
• More efficient search
• complemented with concrete search

t
14
Suggested abstractions

• fewer timepoints in the intervals described by
  the finite domain variables (granularity)
• simpler network - only nodes with significant
  operations such as overturns, meetings and
  reallocation of engines or staff

15
Conclusions

• Constraints and distribution are enabling
  technologies in our application project
• Explicit coordination requires protocols
• Abstract constraint solving maybe useful
• Perhaps tCCP can be useful to bridge gap between
  simulation lt-gt planning