Answer Set Programming (ASP) Encoding

• ASP encoding is the formal representation of combinatorial and epistemic planning problems using logic rules with stable model semantics.
• It enables automated reasoning in dynamic multi-agent systems by integrating modal epistemic operators and DEL-style event updates.
• The approach supports flexible planning through modular representations of actions, fluents, and Kripke structures for complex epistemic scenarios.

Answer Set Programming (ASP) encoding refers to the representation of combinatorial, knowledge-intensive, and dynamic problems as sets of logic rules whose solutions correspond to the answer sets (stable models) of those rules under the stable model semantics. ASP has been widely adopted for automated reasoning, declarative problem solving, multi-agent systems, and epistemic/dynamic planning, with formal links to modal logic, computational logic, and knowledge representation.

1. Formal Underpinnings: Syntax and Semantics

ASP uses logic programming rules of the form

$$a_1 \vee \cdots \vee a_k \leftarrow b_1, \ldots, b_m, \neg c_1, \ldots, \neg c_n$$

where $a_i, b_j, c_l$ are ground atoms built from predicate symbols and constants. The central semantics is the stable model or answer set semantics, which stipulates (via a fixpoint construction) that an answer set is a minimal set of true atoms satisfying all rules, given the “negation as failure” interpretation for default negation $\neg$. This nonmonotonic foundation allows ASP to model commonsense, closed-world, and dynamic domains naturally.

In multi-agent epistemic planning, logic-based encodings are extended with modal epistemic operators $K_i$, $E_\alpha$, and $C_\alpha$ for agent knowledge/belief, group knowledge, and common knowledge, respectively. These are interpreted over Kripke or possibility structures, where worlds encode possible states of the environment and agent relations capture indistinguishability or uncertainty. Epistemic states are therefore higher-order objects—pointed Kripke models $M = (S, \pi, R_1, \ldots, R_n; s)$ as discussed in (Fabiano, 2019)—requiring suitable encodings or representations for use within ASP systems.

2. High-Level ASP Encodings for Epistemic Planning

Multi-agent epistemic planning tasks are mapped into ASP as follows:

• Fluents and Worlds: Ground atoms represent possible assignments of propositional fluents in different worlds. For $f \in \mathcal{F}$ and world $w \in S$, the atom $\text{holds}(f, w)$ may be used.
• Accessibility Relations: Predicates encode agent accessibility: $a_i, b_j, c_l$0 holds if agent $a_i, b_j, c_l$1 cannot distinguish $a_i, b_j, c_l$2 from $a_i, b_j, c_l$3.
• Modal Operators: Knowledge/belief formulas $a_i, b_j, c_l$4, $a_i, b_j, c_l$5, and $a_i, b_j, c_l$6 are compiled to ASP constraints over sets of worlds, enforcing the modal truth conditions within the stable model:
  • $a_i, b_j, c_l$7 at $a_i, b_j, c_l$8 is true iff $a_i, b_j, c_l$9 holds at all $\neg$0 accessible from $\neg$1 via $\neg$2.
  • $\neg$3 is true iff $\neg$4 holds for all $\neg$5 in $\neg$6.
  • $\neg$7 is true iff $\neg$8 holds at $\neg$9 and, recursively, $K_i$0 holds.
• Actions and Event Models: Actions are represented as transition rules between epistemic states. Each action, associated with a DEL-style event model $K_i$1, gives rise to ASP rules and constraints encoding the possibility of applicable events per world, the update of fluents, and the induced changes to agent relations.
• Plan and Goal Encoding: The planning objective is encoded as a sequence or set of actions whose updates transform the initial epistemic state into one satisfying the goal epistemic formula. This is checked via the encoding of modal truth conditions in the resultant state.

In practical encodings, complex Kripke structures may be symbolically represented (e.g., via BDDs or pointer structures) and updated according to the product-update construction, as detailed for DEL-style planning in (Fabiano, 2019).

3. Encoding Dynamic Domains and DEL Updates

The inclusion of DEL-based event models in ASP encoding necessitates the representation of the product update of epistemic models. Given a pointed Kripke model $K_i$2 and an event model $K_i$3, the update yields a new structure $K_i$4:

• Worlds: $K_i$5
• Fluents: $K_i$6 is determined by applying the postcondition of $K_i$7 to $K_i$8.
• Relations: $K_i$9 links $E_\alpha$0 if $E_\alpha$1 and $E_\alpha$2. This construction can be directly encoded within ASP by introducing atoms for pairings of worlds and events and ASP rules that propagate truth assignments and relations subject to pre- and postconditions.

The modular approach allows plugging in user-defined update models, supporting arbitrary epistemic actions including lying, trust manipulation, and private communication, as highlighted in (Fabiano, 2021).

4. ASP Encodings and Multi-Agent Language Compatibility

Recent frameworks aim to ensure that epistemic planning languages—and their ASP encodings—support rich epistemic goals, arbitrary nesting of knowledge, flexible action schemas, and compatibility with other planning formalisms:

• The encoding can parse PDDL-style declarative planning domains extended with epistemic constructs. Agent, fluent, action, and goal declarations are mapped onto the corresponding ASP rules.
• User-defined update models, parameterized by syntactically provided event structures, are encoded to support both fixed ontic and sensing actions as well as plug-in DEL models.
• The ASP encoding must enforce the adopted modal logic's properties—e.g., S5 for knowledge (reflexive, transitive, symmetric relations), KD45 for belief (serial, transitive, Euclidean), and handle common knowledge—thereby ensuring faithful reasoning about agents’ epistemic attitudes.

5. Computational Properties and Solver Architectures

The complexity of propositional epistemic planning with ASP encodings aligns with the underlying modal logic:

• Satisfiability is NP-complete for single-agent S5/KD45, PSPACE-complete for multi-agent S5/KD45, and EXPTIME-complete with common knowledge (Fabiano, 2019).
• Plan existence is EXPSPACE-complete for nonfactual actions with propositional preconditions and undecidable for epistemic preconditions.

State-of-the-art solvers, such as the C++ planner and companion ASP encoding mentioned in (Fabiano, 2021) and (Fabiano, 2019), implement dynamic programming over e-states (epistemic states) with transition function plug-ins, pruning, and support for heuristics such as epistemic planning graphs. The ASP encoding backs symbolic reasoning and transition computation when explicit Kripke structure enumeration is infeasible.

6. Illustrative Example: The Coin Toss Domain

A canonical example encodes agents $E_\alpha$3 and the fluent $E_\alpha$4, with the objective $E_\alpha$5—both agents know the coin’s state. Encoded in ASP:

• Initial state: Two worlds $E_\alpha$6, $E_\alpha$7 true in $E_\alpha$8 only. Both agents' accessibility relations are total (complete ignorance).
• flip action: Non-deterministic, updates the actual world to $E_\alpha$9 or $C_\alpha$0.
• peek$C_\alpha$1: Epistemic action, represented as an event model; only agent $C_\alpha$2 distinguishes outcomes, $C_\alpha$3 does not.
• tell$C_\alpha$4: Announcement action; upon execution, agent $C_\alpha$5 learns the coin state, updating accessibility accordingly.

Stable models represent possible epistemic model evolutions; the correctness of the plan is encoded as an ASP constraint that holds only if, after all actions, $C_\alpha$6 is satisfied in all answer sets. This mirrors the stepwise encoding presented in (Fabiano, 2019).

7. Position in the Research Landscape

ASP encoding for epistemic planning extends classical logic programming to capture multi-agent knowledge dynamics, integrating advances from dynamic epistemic logic, knowledge representation, and planning. The approach described in (Fabiano, 2021) and (Fabiano, 2019) differs from pure PDDL compilation or specialized update mechanisms by supporting arbitrary epistemic updates, unbounded nesting of knowledge modalities, and flexible integration of state representations (Kripke, possibilities, BDDs). This paradigm is suited to domains requiring detailed modeling of information flow, agent perspective, or nonmonotonic knowledge evolution. It provides a rigorous, semantics-preserving route from high-level epistemic planning formalisms to computational implementations via ASP.