Dillon Z. Chen

Dillon Z. Chen, Till Hofmann, Toryn Q. Klassen et al.

Generalised planning (GP) refers to the task of synthesising programs that solve families of related planning problems. We introduce a novel, yet simple method for GP: given a set of training problems, for each problem, compute an optimal plan for each goal atom in some order, perform goal regression on the resulting plans, and lift the corresponding outputs to obtain a set of first-order $\textit{Condition} \rightarrow \textit{Actions}$ rules. The rules collectively constitute a generalised plan that can be executed as is or alternatively be used to prune the planning search space. We formalise and prove the conditions under which our method is guaranteed to learn valid generalised plans and state space pruning axioms for search. Experiments demonstrate significant improvements over state-of-the-art (generalised) planners with respect to the 3 metrics of synthesis cost, planning coverage, and solution quality on various classical and numeric planning domains.

Dillon Z. Chen, Till Hofmann, Toryn Q. Klassen et al.

We tackle the challenge of building embodied AI agents that can reliably solve long-horizon planning problems. Imitation learning from demonstrations has shown itself to be effective in training robots to solve a diversity of complex tasks requiring fine motor control and manipulation over low-level (LL), continuous environments. Yet, it remains a difficult endeavour to generate long-horizon plans from imitation learning alone. In contrast, high-level (HL), symbolic abstractions facilitate efficient and interpretable long-horizon planning. We propose to combine the strengths of LL imitation learning for manipulation and control, and HL symbolic abstractions for long-horizon planning. We realise this idea via \emph{bilevel policies} of the form $(π^{\mathrm{hl}}, π^{\mathrm{ll}})$, consisting of a neural policy $π^{\mathrm{ll}}$ learned from LL demonstrations, and an HL symbolic policy $π^{\mathrm{hl}}$ that is constructed from symbolic abstractions of the LL demonstrations combined with inductive generalisation. We implement these ideas in the BISON system. Experiments on extended MetaWorld benchmarks demonstrate that BISON generalises to long horizons and problems with greater numbers of objects than those solved by VLA and end-to-end methods, and is more time and memory efficient in training and inference. Notably, when ignoring LL execution, BISON's HL policies can solve HL problems with 10,000 relevant objects in under a minute. Project page: https://dillonzchen.github.io/bison

Dillon Z. Chen, Sylvie Thiébaux

Graph learning is naturally well suited for use in symbolic, object-centric planning due to its ability to exploit relational structures exhibited in planning domains and to take as input planning instances with arbitrary numbers of objects. Numeric planning is an extension of symbolic planning in which states may now also exhibit numeric variables. In this work, we propose data-efficient and interpretable machine learning models for learning to solve numeric planning tasks. This involves constructing a new graph kernel for graphs with both continuous and categorical attributes, as well as new optimisation methods for learning heuristic functions for numeric planning. Experiments show that our graph kernels are vastly more efficient and generalise better than graph neural networks for numeric planning, and also yield competitive coverage performance compared to domain-independent numeric planners. Code is available at https://github.com/DillonZChen/goose

Dillon Z. Chen, Sylvie Thiébaux, Felipe Trevizan

We present three novel graph representations of planning tasks suitable for learning domain-independent heuristics using Graph Neural Networks (GNNs) to guide search. In particular, to mitigate the issues caused by large grounded GNNs we present the first method for learning domain-independent heuristics with only the lifted representation of a planning task. We also provide a theoretical analysis of the expressiveness of our models, showing that some are more powerful than STRIPS-HGN, the only other existing model for learning domain-independent heuristics. Our experiments show that our heuristics generalise to much larger problems than those in the training set, vastly surpassing STRIPS-HGN heuristics.

Dillon Z. Chen, Felipe Trevizan, Sylvie Thiébaux

Current approaches for learning for planning have yet to achieve competitive performance against classical planners in several domains, and have poor overall performance. In this work, we construct novel graph representations of lifted planning tasks and use the WL algorithm to generate features from them. These features are used with classical machine learning methods which have up to 2 orders of magnitude fewer parameters and train up to 3 orders of magnitude faster than the state-of-the-art deep learning for planning models. Our novel approach, WL-GOOSE, reliably learns heuristics from scratch and outperforms the $h^{\text{FF}}$ heuristic in a fair competition setting. It also outperforms or ties with LAMA on 4 out of 10 domains on coverage and 7 out of 10 domains on plan quality. WL-GOOSE is the first learning for planning model which achieves these feats. Furthermore, we study the connections between our novel WL feature generation method, previous theoretically flavoured learning architectures, and Description Logic Features for planning.

Dillon Z. Chen, Pulkit Verma, Siddharth Srivastava et al.

Automated decision-making is a fundamental topic that spans multiple sub-disciplines in AI: reinforcement learning (RL), AI planning (AP), foundation models, and operations research, among others. Despite recent efforts to ``bridge the gaps'' between these communities, there remain many insights that have not yet transcended the boundaries. Our goal in this paper is to provide a brief and non-exhaustive primer on ideas well-known in AP, but less so in other sub-disciplines. We do so by introducing the classical AP problem and representation, and extensions that handle uncertainty and time through the Markov Decision Process formalism. Next, we survey state-of-the-art techniques and ideas for solving AP problems, focusing on their ability to exploit problem structure. Lastly, we cover subfields within AP for learning structure from unstructured inputs and learning to generalise to unseen scenarios and situations.

Dillon Z. Chen, Sylvie Thiébaux

Heuristic search is a powerful approach for solving planning problems and numeric planning is no exception. In this paper, we boost the performance of heuristic search for numeric planning with various powerful techniques orthogonal to improving heuristic informedness: numeric novelty heuristics, the Manhattan distance heuristic, and exploring the use of multi-queue search and portfolios for combining heuristics.

Dillon Z. Chen

Weisfeiler-Leman Features (WLFs) are a recently introduced classical machine learning tool for learning to plan and search. They have been shown to be both theoretically and empirically superior to existing deep learning approaches for learning value functions for search in symbolic planning. In this paper, we introduce new WLF hyperparameters and study their various tradeoffs and effects. We utilise the efficiency of WLFs and run planning experiments on single core CPUs with a sample size of 1,000,000 to understand the effect of hyperparameters on training and planning. Our experimental analysis show that there is a robust and best set of hyperparameters for WLFs across the tested planning domains. We find that the best WLF hyperparameters for learning heuristic functions minimise execution time rather than maximise model expressivity. We further statistically analyse and observe no significant correlation between training and planning metrics.

Dillon Z. Chen, Johannes Zenn, Tristan Cinquin et al.

We study the usage of language models (LMs) for planning over world models specified in the Planning Domain Definition Language (PDDL). We prompt LMs to generate Python programs that serve as generalised policies for solving PDDL problems from a given domain. Notably, our approach synthesises policies that are provably sound relative to the PDDL domain without reliance on external verifiers. We conduct experiments on competition benchmarks which show that our policies can solve more PDDL problems than PDDL planners and recent LM approaches within a fixed time and memory constraint. Our approach manifests in the LMPlan planner which can solve planning problems with several hundreds of relevant objects. Surprisingly, we observe that LMs used in our framework sometimes plan more effectively over PDDL problems written in meaningless symbols in place of natural language; e.g. rewriting (at dog kitchen) as (p2 o1 o3). This finding challenges hypotheses that LMs reason over word semantics and memorise solutions from its training corpus, and is worth further exploration.

Dillon Z. Chen

Novelty heuristics aid heuristic search by exploring states that exhibit novel atoms. However, novelty heuristics are not symmetry invariant and hence may sometimes lead to redundant exploration. In this preliminary report, we propose to use Weisfeiler-Leman Features for planning (WLFs) in place of atoms for detecting novelty. WLFs are recently introduced features for learning domain-dependent heuristics for generalised planning problems. We explore an unsupervised usage of WLFs for synthesising lifted, domain-independent novelty heuristics that are invariant to symmetric states. Experiments on the classical International Planning Competition and Hard To Ground benchmark suites yield promising results for novelty heuristics synthesised from WLFs.

Dillon Z. Chen

Relational Graph Neural Networks (R-GNNs) are a GNN-based approach for learning value functions that can generalise to unseen problems from a given planning domain. R-GNNs were theoretically motivated by the well known connection between the expressive power of GNNs and $C_2$, first-order logic with two variables and counting. In the context of planning, $C_2$ features refer to the set of formulae in $C_2$ with relations defined by the unary and binary predicates of a planning domain. Some planning domains exhibit optimal value functions that can be decomposed as arithmetic expressions of $C_2$ features. We show that, contrary to empirical results, R-GNNs cannot learn value functions defined by $C_2$ features. We also identify prior GNN architectures for planning that may better learn value functions defined by $C_2$ features.

Dillon Z. Chen, Mingyu Hao, Sylvie Thiébaux et al.

Graph learning is naturally well suited for use in planning due to its ability to exploit relational structures exhibited in planning domains and to take as input planning instances with arbitrary number of objects. In this paper, we study the usage of graph learning for planning thus far by studying the theoretical and empirical effects on learning and planning performance of (1) graph representations of planning tasks, (2) graph learning architectures, and (3) optimisation formulations for learning. Our studies accumulate in the GOOSE framework which learns domain knowledge from small planning tasks in order to scale up to much larger planning tasks. In this paper, we also highlight and propose the 5 open challenges in the general Learning for Planning field that we believe need to be addressed for advancing the state-of-the-art.

Dillon Z. Chen

Scalable learning for planning research generally involves juggling between different programming languages for handling learning and planning modules effectively. Interpreted languages such as Python are commonly used for learning routines due to their ease of use and the abundance of highly maintained learning libraries they exhibit, while compiled languages such as C++ are used for planning routines due to their optimised resource usage. Motivated by the need for tools for developing scalable learning planners, we introduce WLPlan, a C++ package with Python bindings which implements recent promising work for automatically generating relational features of planning tasks. Such features can be used for any downstream routine, such as learning domain control knowledge or probing and understanding planning tasks. More specifically, WLPlan provides functionality for (1) transforming planning tasks into graphs, and (2) embedding planning graphs into feature vectors via graph kernels. The source code and instructions for the installation and usage of WLPlan are available at tinyurl.com/42kymswc