Automated planning is one of the oldest problem areas in Artificial
Intelligence and as such has motivated numerous implementations, but many
of these programs are poorly understood. The major objective of this
project is developing precise algorithms for a wide class of planning
problems, investigating the optimizations and approximations necessary to
apply our algorithms to large, realistic problems, and integrating planning
and execution strategies. Some issues addressed include:
- Efficiently maintaining soundness and completeness with an
expressive action language (conditional effects, universally quantified
effects and goals).
- Implementation of metric functions and minimization goals using
ideas from linear programming and CLP(R).
- Temporal planning with simultaneous actions, exogenous event,
continuous change, and deadline goals.
- Planning with a probabilistic model (i.e., with actions whose
effects are described in terms of conditional probability).
- Planning with incomplete information (deliberative and reactive
control of sensing).
- Integrated execution strategies: conditional planning vs
interleaved planning and execution.
- Efficiently controlling inference through domain axioms and search
control techniques.
- Application of planning to software agents.
Principal Investigator: Weld