When LLMs exhibit uneven performance across planning tasks, these gaps are often attributed to task difficulty. We argue that this explanation is incomplete, as task-level variation may reflect distinct latent planning competencies rather than differences along a single ability spectrum. We study this question on ACPBench-Hard by evaluating multiple LLM families under varying test-time reasoning budgets and applying a multidimensional item response theory model to uncover the latent competency structure underlying LLM planning. The analysis reveals two principal dimensions that shape planning performance: operational reasoning, the ability to evaluate local action applicability and immediate state transitions, and structural enumeration, the ability to reason about goal reachability and landmark structure. Operational reasoning improving under model scaling and longer reasoning traces, while structural enumeration remains comparatively insensitive. Our findings motivate competency-level evaluation of LLM planning, shifting the focus from whether models improve overall to which planning competencies improve, under what conditions, and why.
Prior work suggests that language models, while trained on next token prediction, show implicit planning behavior: they may select the next token in preparation to a predicted future token, such as a likely rhyming word, as supported by a prior qualitative study of Claude 3.5 Haiku using a cross-layer transcoder. We propose much simpler techniques for assessing implicit planning in language models. With case studies on rhyme poetry generation and question answering, we demonstrate that our methodology easily scales to many models. Across models, we find that the generated rhyme (e.g. "-ight") or answer to a question ("whale") can be manipulated by steering at the end of the preceding line with a vector, affecting the generation of intermediate tokens leading up to the rhyme or answer word. We show that implicit planning is a universal mechanism, present in smaller models than previously thought, starting from 1B parameters. Our methodology offers a widely applicable direct way to study implicit planning abilities of LLMs. More broadly, understanding planning abilities of language models can inform decisions in AI safety and control.
Large Language Models enable flexible natural-language planning but remain unreliable in determinism-critical domains due to their probabilistic nature. This limitation is especially problematic in running planning, where violating safety rules can lead to safety risks. We propose SafeRun, a framework for deterministic LLM-based planning via a decoupled architecture. SafeRun separates soft interpretation by an LLM from hard constraint enforcement by a deterministic solver, ensuring strict safety constraints while preserving natural-language flexibility. To validate SafeRun, we build a comprehensive benchmark for running planning under realistic physiological and safety constraints. Experiments across five LLMs show that SafeRun achieves 100\% safety score (vs.\ 79.1\% PE average and 97.6\% CodeAct average) while maintaining competitive instruction-following scores. The SafeRun benchmark is publicly available at \href{https://huggingface.co/datasets/zzp-seeker/SafeRun-RunPlanning-Benchmark}{huggingface}.
Planning safe paths in 3D workspace for high DoF robotic systems, such as manipulators, is a challenging problem, especially when the environment is populated with the dynamic obstacles that need to be avoided. In this case the time dimension should be taken into account that further increases the complexity of planning. To mitigate this issue we suggest to combine safe-interval path planning (a prominent technique in heuristic search) with the randomized planning, specifically, with the bidirectional rapidly-exploring random trees (RRT-Connect) - a fast and efficient algorithm for high-dimensional planning. Leveraging a dedicated technique of fast computation of the safe intervals we end up with an efficient planner dubbed SI-RRT. We compare it with the state of the art and show that SI-RRT consistently outperforms the competitors both in runtime and solution cost. Our implementation of SI-RRT is publicly available at https://github.com/PathPlanning/ManipulationPlanning-SI-RRT
In the real world, planning is often challenged by distribution shifts. As such, a model of the environment obtained under one set of conditions may no longer remain valid as the distribution of states or the environment dynamics change, which in turn causes previously learned strategies to fail. In this work, we propose a theoretical framework for planning under partial observability using Partially Observable Markov Decision Processes (POMDPs) formulated using causal knowledge. By representing shifts in the environment as interventions on this causal POMDP, the framework enables evaluating plans under hypothesized changes and actively identifying which components of the environment have been altered. We show how to maintain and update a belief over both the latent state and the underlying domain, and we prove that the value function remains piecewise linear and convex (PWLC) in this augmented belief space. Preservation of PWLC under distribution shifts has the advantage of maintaining the tractability of planning via $α$-vector-based POMDP methods.
Radiotherapy (RT) planning is complex, subjective, and time-intensive. Advances with artificial intelligence (AI) promise to improve its precision and efficiency, but progress is often limited by the scarcity of large, standardized datasets. To address this, we introduce the Automated Iterative RT Planning (AIRTP) system, a scalable solution for generating high-quality treatment plans. This scalable solution is designed to generate substantial volumes of consistently high-quality treatment plans, overcoming a key obstacle in the advancement of AI-driven RT planning. Our AIRTP pipeline adheres to clinical guidelines and automates essential steps, including organ-at-risk (OAR) contouring, helper structure creation, beam setup, optimization, and plan quality improvement, using AI integrated with RT planning software like Varian Eclipse. Furthermore, a novel approach for determining optimization parameters to reproduce 3D dose distributions, i.e. a method to convert dose predictions to deliverable treatment plans constrained by machine limitations is proposed. A comparative analysis of plan quality reveals that our automated pipeline produces treatment plans of quality comparable to th
Long-horizon planning for robot manipulation is a challenging problem that requires reasoning about the effects of a sequence of actions on a physical 3D scene. While traditional task planning methods are shown to be effective for long-horizon manipulation, they require discretizing the continuous state and action space into symbolic descriptions of objects, object relationships, and actions. Instead, we propose a hybrid learning-and-planning approach that leverages learned models as domain-specific priors to guide search in high-dimensional continuous action spaces. We introduce SPOT: Search over Point cloud Object Transformations, which plans by searching for a sequence of transformations from an initial scene point cloud to a goal-satisfying point cloud. SPOT samples candidate actions from learned suggesters that operate on partially observed point clouds, eliminating the need to discretize actions or object relationships. We evaluate SPOT on multi-object rearrangement tasks, reporting task planning success and task execution success in both simulation and real-world environments. Our experiments show that SPOT generates successful plans and outperforms a policy-learning approac
Ontologies are known for their ability to organize rich metadata, support the identification of novel insights via semantic queries, and promote reuse. In this paper, we consider the problem of automated planning, where the objective is to find a sequence of actions that will move an agent from an initial state of the world to a desired goal state. We hypothesize that given a large number of available planners and diverse planning domains; they carry essential information that can be leveraged to identify suitable planners and improve their performance for a domain. We use data on planning domains and planners from the International Planning Competition (IPC) to construct a planning ontology and demonstrate via experiments in two use cases that the ontology can lead to the selection of promising planners and improving their performance using macros - a form of action ordering constraints extracted from planning ontology. We also make the planning ontology and associated resources available to the community to promote further research.
Recent advances in Large Language Models (LLMs) are fostering their integration into several reasoning-related fields, including Automated Planning (AP). However, their integration into Hierarchical Planning (HP), a subfield of AP that leverages hierarchical knowledge to enhance planning performance, remains largely unexplored. In this preliminary work, we propose a roadmap to address this gap and harness the potential of LLMs for HP. To this end, we present a taxonomy of integration methods, exploring how LLMs can be utilized within the HP life cycle. Additionally, we provide a benchmark with a standardized dataset for evaluating the performance of future LLM-based HP approaches, and present initial results for a state-of-the-art HP planner and LLM planner. As expected, the latter exhibits limited performance (3\% correct plans, and none with a correct hierarchical decomposition) but serves as a valuable baseline for future approaches.
We introduce lower-bound certificates for classical planning tasks, which can be used to prove the unsolvability of a task or the optimality of a plan in a way that can be verified by an independent third party. We describe a general framework for generating lower-bound certificates based on pseudo-Boolean constraints, which is agnostic to the planning algorithm used. As a case study, we show how to modify the $A^{*}$ algorithm to produce proofs of optimality with modest overhead, using pattern database heuristics and $h^\textit{max}$ as concrete examples. The same proof logging approach works for any heuristic whose inferences can be efficiently expressed as reasoning over pseudo-Boolean constraints.
In this work, we argue that large language models (LLMs), though trained to predict only the next token, exhibit emergent planning behaviors: $\textbf{their hidden representations encode future outputs beyond the next token}$. Through simple probing, we demonstrate that LLM prompt representations encode global attributes of their entire responses, including $\textit{structure attributes}$ (e.g., response length, reasoning steps), $\textit{content attributes}$ (e.g., character choices in storywriting, multiple-choice answers at the end of response), and $\textit{behavior attributes}$ (e.g., answer confidence, factual consistency). In addition to identifying response planning, we explore how it scales with model size across tasks and how it evolves during generation. The findings that LLMs plan ahead for the future in their hidden representations suggest potential applications for improving transparency and generation control.
Motivated by the commitments from the Talmud in Judaism, we consider the family planning rules which require a couple to get children till certain numbers of boys and girls are reached. For example, the rabbinical school of Beit Hillel says that one boy and one girl are necessary, whereas Beit Shammai urges for two boys. Surprisingly enough, although the corresponding average family sizes differ in both cases, the gender ratios remain constant. We show more that for any family planning rule the gender ratio is equal to the birth odds. The proof of this result is given by using different mathematical techniques, such as induction principle, Doob's optional-stopping theorem, and brute-force. We conclude that, despite possible asymmetries in the religiously motivated family planning rules, they discriminate neither boys nor girls.
Recent advancements in Large Language Models (LLMs) have showcased their ability to perform complex reasoning tasks, but their effectiveness in planning remains underexplored. In this study, we evaluate the planning capabilities of OpenAI's o1 models across a variety of benchmark tasks, focusing on three key aspects: feasibility, optimality, and generalizability. Through empirical evaluations on constraint-heavy tasks (e.g., $\textit{Barman}$, $\textit{Tyreworld}$) and spatially complex environments (e.g., $\textit{Termes}$, $\textit{Floortile}$), we highlight o1-preview's strengths in self-evaluation and constraint-following, while also identifying bottlenecks in decision-making and memory management, particularly in tasks requiring robust spatial reasoning. Our results reveal that o1-preview outperforms GPT-4 in adhering to task constraints and managing state transitions in structured environments. However, the model often generates suboptimal solutions with redundant actions and struggles to generalize effectively in spatially complex tasks. This pilot study provides foundational insights into the planning limitations of LLMs, offering key directions for future research on impro
Classical planners are powerful systems, but modeling tasks in input formats such as PDDL is tedious and error-prone. In contrast, planning with Large Language Models (LLMs) allows for almost any input text, but offers no guarantees on plan quality or even soundness. In an attempt to merge the best of these two approaches, some work has begun to use LLMs to automate parts of the PDDL creation process. However, these methods still require various degrees of expert input or domain-specific adaptations. We present NL2Plan, the first fully automatic system for generating complete PDDL tasks from minimal natural language descriptions. NL2Plan uses an LLM to incrementally extract the necessary information from the short text input before creating a complete PDDL description of both the domain and the problem which is finally solved by a classical planner. We evaluate NL2Plan on seven planning domains, five of which are novel and thus not in the LLM training data, and find that NL2Plan outperforms directly generating the files with an LLM+validator combination. As such, NL2Plan is a powerful tool for assistive PDDL modeling and a step towards solving natural language planning task with in
Automated planning is a major topic of research in artificial intelligence, and enjoys a long and distinguished history. The classical paradigm assumes a distinguished initial state, comprised of a set of facts, and is defined over a set of actions which change that state in one way or another. Planning in many real-world settings, however, is much more involved: an agent's knowledge is almost never simply a set of facts that are true, and actions that the agent intends to execute never operate the way they are supposed to. Thus, probabilistic planning attempts to incorporate stochastic models directly into the planning process. In this article, we briefly report on probabilistic planning through the lens of probabilistic programming: a programming paradigm that aims to ease the specification of structured probability distributions. In particular, we provide an overview of the features of two systems, HYPE and ALLEGRO, which emphasise different strengths of probabilistic programming that are particularly useful for complex modelling issues raised in probabilistic planning. Among other things, with these systems, one can instantiate planning problems with growing and shrinking state
Randomized sampling based algorithms are widely used in robot motion planning due to the problem's intractability, and are experimentally effective on a wide range of problem instances. Most variants do not sample uniformly at random, and instead bias their sampling using various heuristics for determining which samples will provide more information, or are more likely to participate in the final solution. In this work, we define the \emph{motion planning guiding space}, which encapsulates many seemingly distinct prior works under the same framework. In addition, we suggest an information theoretic method to evaluate guided planning which places the focus on the quality of the resulting biased sampling. Finally, we analyze several motion planning algorithms in order to demonstrate the applicability of our definition and its evaluation.
Real world applications as in industry and robotics need modelling rich and diverse automated planning problems. Their resolution usually requires coordinated and concurrent action execution. In several cases, these problems are naturally decomposed in a hierarchical way and expressed by a Hierarchical Task Network (HTN) formalism. HDDL, a hierarchical extension of the Planning Domain Definition Language (PDDL), unlike PDDL 2.1 does not allow to represent planning problems with numerical and temporal constraints, which are essential for real world applications. We propose to fill the gap between HDDL and these operational needs and to extend HDDL by taking inspiration from PDDL 2.1 in order to express numerical and temporal expressions. This paper opens discussions on the semantics and the syntax needed for a future HDDL 2.1 extension.
To make effective decisions in novel environments with long-horizon goals, it is crucial to engage in hierarchical reasoning across spatial and temporal scales. This entails planning abstract subgoal sequences, visually reasoning about the underlying plans, and executing actions in accordance with the devised plan through visual-motor control. We propose Compositional Foundation Models for Hierarchical Planning (HiP), a foundation model which leverages multiple expert foundation model trained on language, vision and action data individually jointly together to solve long-horizon tasks. We use a large language model to construct symbolic plans that are grounded in the environment through a large video diffusion model. Generated video plans are then grounded to visual-motor control, through an inverse dynamics model that infers actions from generated videos. To enable effective reasoning within this hierarchy, we enforce consistency between the models via iterative refinement. We illustrate the efficacy and adaptability of our approach in three different long-horizon table-top manipulation tasks.
This paper explores the application of automated planning to automated theorem proving, which is a branch of automated reasoning concerned with the development of algorithms and computer programs to construct mathematical proofs. In particular, we investigate the use of planning to construct elementary proofs in abstract algebra, which provides a rigorous and axiomatic framework for studying algebraic structures such as groups, rings, fields, and modules. We implement basic implications, equalities, and rules in both deterministic and non-deterministic domains to model commutative rings and deduce elementary results about them. The success of this initial implementation suggests that the well-established techniques seen in automated planning are applicable to the relatively newer field of automated theorem proving. Likewise, automated theorem proving provides a new, challenging domain for automated planning.
We propose a new benchmark for planning tasks based on the Minecraft game. Our benchmark contains 45 tasks overall, but also provides support for creating both propositional and numeric instances of new Minecraft tasks automatically. We benchmark numeric and propositional planning systems on these tasks, with results demonstrating that state-of-the-art planners are currently incapable of dealing with many of the challenges advanced by our new benchmark, such as scaling to instances with thousands of objects. Based on these results, we identify areas of improvement for future planners. Our framework is made available at https://github.com/IretonLiu/mine-pddl/.