Temporal Planning Unsolvability Certification

• The paper introduces a rigorous method for certifying the unsolvability of temporal planning instances using quantified constraint frameworks and automata-theoretic representations.
• It establishes tight complexity bounds for strong, weak, and dynamic controllability while leveraging recursive simulation and hybrid linear programming approaches.
• The study integrates formal verification and explainable AI techniques, including model checking, hierarchical abstraction, and waypoint extraction, to enhance trust in unsolvability certificates.

Unsolvability certification in temporal planning constitutes a rigorous methodology for providing formal evidence that no feasible plan satisfying the imposed temporal constraints exists for a given planning instance. This topic envelops complexity-theoretic guarantees, expressive modeling features, sound and efficient certification algorithms, and context-sensitive explanations. Approaches range from the use of quantified constraint frameworks, conversion to automata-theoretic representations for verification, and advanced methods for explainable unsolvability grounded in task abstraction or symbolic reachability. The field leverages both decision procedures and explanation paradigms—integrating complexity results, algorithmic soundness, and formal verification to establish trust in unsolvability certificates.

1. Computational Complexity of Unsolvability in Temporal Networks

Controllability checking of temporal networks—especially those with conditions, disjunctions, and temporal uncertainty—admits tight complexity class bounds, which directly inform unsolvability certification (Bhargava et al., 2019). Three controllability notions are crucial:

• Strong controllability: In basic models (STN, STNU, Conditional STNU), decidable in polynomial time; with disjunctions, becomes NP-complete (e.g., TCSPU).
• Weak controllability: Elevates the problem to coNP-complete for CSTNUs and to higher levels in the polynomial hierarchy when disjunctions and uncertainty interact.
• Dynamic controllability: For expressive models involving both conditions and disjunctions (DTNUs, CDTNUs), recognition of unsolvability is PSPACE-complete, shown by reductions from TQBF.

For unsolvability certification, these completeness results indicate that certifiers must quantify over scheduler strategies and all possible behaviors of uncontrollable events, commonly formalized as

$$\exists \mathbf{x} \forall \mathbf{y}: \bigwedge_i(A_i \cdot [\mathbf{x}\ \mathbf{y}]^\top \leq b_i)$$

where $\mathbf{x}$ are controllable assignments and $\mathbf{y}$ correspond to nature’s choices. The dual alternation of quantifiers precisely encodes the logical structure of unsolvability proofs in temporal planning.

2. Expressiveness and Modeling Features Impacting Certification

Temporal planning models are increasingly expressive, featuring:

• Conditional constraints: Propositional labels activating constraints contingent on external information, leading to quantification over Boolean space.
• Disjunctive constraints: Simple disjunctions (TCSPs) vs. full disjunctions (DTNs), the latter involving arbitrary tuples and inducing a combinatorial explosion in the number of possibilities for certification algorithms.
• Temporal uncertainty: Uncontrollable contingent durations, transforming the verification into a two-player game against an adversarial environment.

These features introduce substantial complexity into the structure of unsolvability certificates, as certification must demonstrate the absence of any scheduler strategy robust to every realization of conditions and uncertainty. The inherent alternation of quantifiers and the explosion in feasible scenarios force certificate algorithms to operate in high-complexity regimes, both in theory and in practice (Bhargava et al., 2019).

3. Algorithmic Frameworks for Unsolvability Certification

Algorithmic advances for unsolvability certification rely on quantified constraint satisfaction, recursion, and symbolic verification:

• Quantified linear systems: Temporal planning problems are transformed into systems of conditional linear inequalities, typically formulated as

$$\exists \mathbf{x} \forall \mathbf{y}\ \forall \Psi:\ \bigwedge_i(\psi_i \rightarrow (a_i \cdot [\mathbf{x}\ \mathbf{y}]^\top \leq b_i))$$

where $\Psi$ is the set of condition observations.

• Recursive simulation: For dynamic controllability, recursive algorithms simulate every possible scheduler–nature interaction using polynomial space, enabling both the exhaustive search for a dynamic execution policy and the generation of counterexamples certifying unsolvability.
• Hybrid approaches: Integration of linear programming subroutines verifies the absence of feasible scheduler assignments for fixed uncontrollable choices.

End-to-end, these approaches enable refutation-based certification—documenting the precise structure of failed branches, and yielding robust artifacts for verification tools grounded in theoretical guarantees of space complexity (Bhargava et al., 2019).

4. Verification via Formal Methods and Model Checking

Formally verified unsolvability certification offers high assurance in critical applications. The principal methodology proceeds as follows (Wang et al., 11 Oct 2025):

• Timed Automata Network Encoding: Temporal planning problems are encoded as networks of timed automata, with each durative action represented by a pair of snap transitions (start/finish), state and lock variables recording the evolution of propositions, and clocks enforcing duration and separation constraints. For example, a transition representing action start will have guards as:

$\ell_a \leq c_{a_\vdash} < u_a$

ensuring compliance with temporal requirements.

• Model Checking: An efficient (possibly unverified) model checker evaluates reachability properties, typically using TCTL formula EF(goal) to detect whether the goal location is attainable in the automata network.
• Certificate Checking: The output certificate—trace or invariant barrier—from model checking is then subject to verification in a formally verified certificate checker (e.g., in Isabelle/HOL), ensuring mathematical correctness of the unsolvability claim.
• Formal Verification of Encoding: The encoding and certificate checker are formally verified, ensuring every valid plan in the temporal planning problem is simulated exactly in the timed automata model.

This two-staged approach combines the scalability and rapid search capabilities of external solvers with the rigorous trustworthiness of proof-based certificate verification (Wang et al., 11 Oct 2025).

5. Explanatory and Diagnostic Frameworks

Recent research pivots toward human-understandable explanations for unsolvability, complementing raw certification:

• Hierarchical abstraction: Planning models are projected onto abstraction lattices, with unsolvability revealed at minimal abstraction levels, and explanatory fluents and unachievable landmarks pinpointing the obstruction (Sreedharan et al., 2019).
• Inevitable waypoint identification: Every feasible plan traverses a core chain of waypoints, determined by computing the longest common subsequence (LCS) of all bounded-length paths in the abstract planning graph; unsolvability is explained by the first unreachable waypoint, detected by symbolic reachability analysis (Sarwar et al., 22 Apr 2025).
• Constraint satisfaction propagation: Feasibility tensors encode the probability of subgoal satisfaction; unsolvability is certified when the maximal feasibility along any goal sequence drops to zero, allowing cycle-level diagnosis of unsolvable subproblems (Ringstrom et al., 2019).
• Certification workflow summary:

| Approach | Key Principle | Type of Explanation | |-----------------------|-------------------------------|-----------------------------------| | Quantified constraints| Alternating quantifiers | Refutation/counterexample | | Timed automata | Formal reachability checking | Unsatisfiable certificate | | Hierarchical lattice | Landmarks, abstraction | Human-interpretable subgoal | | Waypoint chain | Path LCS, symbolic checking | Earliest unreachable waypoint |

These frameworks foster interpretability and facilitate not only formal verification, but also diagnosis and potential relaxation of constraints for future plan feasibility assessment.

6. Implications and Best Practices

Certification of unsolvability in temporal planning has profound implications:

• Model design trade-offs: Inclusion of expressive features (disjunctions, uncertainty, conditional activation) may escalate verification complexity, which informs the modeling choices for scalable certification.
• Tool design: PSPACE-computability ensures practical algorithms may have polynomial space footprints, but time requirements scale exponentially in the worst case; verification and certification toolchains must accommodate this.
• Trustworthiness in critical settings: Formal verification, using tools such as Isabelle/HOL, can ensure high-assurance unsolvability certification in domains such as robotics, manufacturing, or high-stakes logistics (Wang et al., 11 Oct 2025, Abdulaziz et al., 2022).
• Explainability: Methods for landmark and waypoint extraction, as well as modular constraint propagation, advance the field toward explainable AI planning, increasing transparency for operators and system stakeholders (Sarwar et al., 22 Apr 2025, Sreedharan et al., 2019).

Current and emerging techniques facilitate robust, interpretable, and practically efficient unsolvability certification for temporal planning, offering methodological rigor across algorithmic, formal, and explanatory dimensions.