goal reachability inference

This work addresses a critical limitation in existing automated scientific discovery systems, which rely on myopic information-gain strategies and fail to evaluate the long-term value of constructive actions—such as developing new instruments—in problems requiring chains of capabilities to achieve a goal. The authors formalize goal-directed scientific discovery as a stochastic shortest path problem in belief space, where constructive experiments dynamically expand the action space. They propose CG-Plan, an incremental replanning algorithm that integrates a capability-aware heuristic combining capability acquisition (h_cap) and experimental progress (h_exp). Theoretical analysis introduces “capability gating” as a novel dimension of problem hardness, proving that any fixed-horizon myopic planner suffers either unbounded approximation ratios or incompleteness in such settings. Experiments demonstrate that CG-Plan substantially outperforms myopic baselines in capability-gated scenarios, with consistent performance advantages across all fixed planning horizons.

Existing target identification design methods incur high computational overhead and critically rely on the assumption of optimal surrogate decision-making, rendering them ill-suited to real-world suboptimal human behavior and complex environments. To address this, we propose a data-driven framework compatible with general behavioral models, which— for the first time—integrates machine learning with constraint-aware gradient optimization to enable adaptive design of decision environments. Our approach constructs a differentiable predictive model based on the Worst-Case Deviation (WCD) metric, supporting flexible resource budgets and explicit modeling of non-optimal strategies. Simulations demonstrate significant WCD reduction and improved runtime efficiency. Human-subject experiments further confirm that our method effectively guides real decision-makers toward faster and more accurate target identification. By relaxing the restrictive optimality assumption, this work extends the applicability of target identification design to practical human–machine collaborative settings.

To address the challenge of explaining infeasibility in hybrid planning, this paper proposes an attribution method based on *inevitable waypoints*—a subset of states that all feasible trajectories must visit yet are provably unreachable. We innovatively formulate inevitable waypoint identification as a Longest Common Subsequence (LCS) optimization problem, integrated with symbolic reachability analysis to precisely pinpoint the earliest point of failure. Unlike prior approaches that attribute infeasibility to local constraint conflicts, our method systematically identifies semantic, universally applicable obstacle waypoints as the root cause. Evaluated on multiple infeasible hybrid planning instances, the approach generates concise, human-readable explanations that accurately expose fundamental obstacles, thereby significantly enhancing the explainability and debugging efficiency of AI planning systems.

This paper addresses the measurability of “goal-directedness” in complex agents, identifying fundamental conceptual ambiguities and formalization challenges in both behavioral observation and internal-state probing—the two dominant methodological approaches. Methodologically, it integrates behavioral analysis, mechanistic explanation, and formal modeling to systematically examine the implicit assumptions and inherent limitations of behaviorist versus mechanist paradigms in goal attribution. The primary contribution is a rigorous demonstration that goal-directedness lacks objective quantifiability; instead, it is an observer-dependent, emergent property arising dynamically within multi-agent interactions. Consequently, the paper proposes a novel, non-reductionist, de-intrinsicized modeling framework for goals—one that eschews internal mental-state commitments and prioritizes relational, interactional structure. This conceptual reframing provides foundational groundwork for advancing AI interpretability, value alignment, and agent evaluation.

Traditional planning models goals as deterministic state subsets, failing to capture goal uncertainty arising from noisy perception, learning generalization, and other real-world uncertainties. Method: This paper proposes a distribution-to-distribution planning framework that directly represents both goals and current states as probability distributions, explicitly accommodating dynamic environmental uncertainty. It introduces goal distributions as fundamental planning primitives, integrates unscented transformation for nonlinear probabilistic kinematics modeling, and employs cross-entropy optimization to minimize the KL divergence between predicted and target distributions—enabling end-to-end uncertainty-aware planning. Contribution/Results: The framework unifies several classical goal cost functions as special cases. Extensive simulations demonstrate strong robustness against state disturbances, model mismatch, and data-driven goal uncertainty. Moreover, it significantly improves task success reliability under multimodal, sparse, and constrained goal distributions.

This study challenges the prevailing assumption that step-by-step monitoring is essential for adaptive performance in data-intensive tasks by systematically investigating planning horizon as an independent variable. Through controlled experiments, the authors compare full-horizon (FH) planning against single-horizon (SH) planning in knowledge-base question answering and multi-hop reasoning tasks. The results demonstrate that FH planning, augmented with on-demand replanning, achieves accuracy comparable to SH planning across varying task depths, breadths, and tool robustness conditions—while reducing token consumption by a factor of 2–3. These findings question the necessity of continuous, fine-grained monitoring in structured data tasks and suggest that broader planning horizons can yield substantial efficiency gains without compromising performance.

This study addresses the failure of long-horizon planning in hierarchical frameworks caused by macro-level predictors generating physically infeasible subgoals. To overcome this, we propose Metro-WM, a framework that introduces a novel graph-planning mechanism based on empirical frame retrieval. Specifically, Metro-WM employs a joint embedding architecture to extract real states from offline expert demonstrations, constructing a graph structure for path search rather than relying on unconstrained latent vector generation. This approach fundamentally ensures the physical feasibility and execution robustness of subgoals. Experimental results demonstrate that Metro-WM improves success rates on long-horizon tasks by 37.33% and accelerates planning speed by 10.9×, significantly reducing computational overhead while discovering shorter paths than those provided in the demonstrations.

This study addresses the disconnect between comprehension and execution in large language model (LLM) agents, which frequently leads to falsely reported task completion. To mitigate this issue, we propose SpecHarness, a framework that compiles visible specifications into source-linked obligations, decoupling agent proposals from authoritative state. Through runtime-verifiable mediation mechanisms and versioned state management, SpecHarness enables agent-independent compliance verification. Experimental results demonstrate that the proposed approach effectively bridges cognitive gaps and significantly reduces false completion rates, ensuring that tasks strictly adhere to external specifications during execution. Ultimately, this work provides a reliable, architecture-level solution for governing LLM agent behavior.

This work challenges the common assumption that poor long-horizon planning stems primarily from inaccurate world model predictions, demonstrating instead that suboptimal planner objective functions are often the true bottleneck. Analyzing latent-variable world models in a TwoRoom environment, the authors show that these models accurately encode distant future states, yet conventional squared Euclidean distance objectives in latent space impede effective planning. By simply replacing the planning objective—without retraining the model—they boost task success rates from 26.0% to 98.0% under a 100-step horizon, and achieve 92.0% success using only one-third of the original computational budget. Through cross-entropy method (CEM) planning, latent-space geometric probing, and multi-objective comparisons, the study reveals that maximal planning performance does not require maximal prediction accuracy; rather, it hinges on aligning the objective function with the underlying task structure.

This work addresses the challenge of achieving efficient goal-directed planning in vision-based world models without relying on computationally expensive online search. It proposes amortizing the planning process into a lightweight inverse dynamics mapping in latent space, leveraging the smooth and uniform geometric structure of a pretrained world model to directly predict actions from the current state, goal state, and remaining time steps. By shifting the planning burden from online optimization to learned inference, this approach reveals that structured latent spaces can intrinsically encode the local geometry necessary for planning. Combining LeWorldModel, latent-space geometric regularization, and a goal-conditioned inverse dynamics model (GC-IDM), the method achieves controller performance that matches or exceeds Cross-Entropy Method (CEM) in seven out of eight settings across four benchmark environments, while reducing per-decision computational cost by 100–130×.