PKS: Knowledge-Based Planning with Incomplete Information and Sensing

PKS (Planning with Knowledge and Sensing) is a “knowledge-level” planner that is able to construct conditional plans in the presence of incomplete knowledge and sensing (Bacchus and Petrick 1998; Petrick and Bacchus 2002; 2004). The key idea of this approach is to represent the agent’s knowledge state with a first-order language, and to represent actions by their effects on the agent’s knowledge, rather than by their effects on the environment. Since general reasoning in such a rich language is impractical, PKS employs a restricted subset of the language and a limited amount of inference in that subset. As a result, PKS includes non-propositional features, such as functions and variables. The knowledge-based approach contrasts some of the alternate trends that have concentrated on propositional representations over which complete reasoning is feasible. Such works often represent the set of all possible worlds (i.e., the set of all states compatible with the agent’s incomplete knowledge) using various techniques (e.g., BDDs, Graphplan-like structures, or clausal representations). These techniques yield planning systems that are able to generate plans requiring complex combinatorial reasoning. By representing problems at the knowledge level, PKS can generate plans that are often quite “natural” and have a simple structure. Furthermore, PKS can often “abstract” away from some of the irrelevant distinctions that occur at the world level. Compared to the possible-worlds approaches, this higher-level representation is richer, but the inferences it supports are weaker. Nevertheless, PKS is able to solve problems that cannot be solved by alternate approaches. We briefly describe some of the important components we have implemented in the PKS system.