Hierarchical Task Network (HTN) planners generate plans using a decomposition process with extra domain knowledge to guide search towards a planning task. Domain experts develop such domain knowledge through recipes of how to decompose higher level tasks, specifying which tasks can be decomposed and under what conditions. In most realistic domains, such recipes contain recursions, i.e., tasks that can be decomposed into other tasks that contain the original task. Such domains require that either the domain expert tailor such domain knowledge to the specific HTN planning algorithm, or an algorithm that can search efficiently using such domain knowledge. By leveraging a three-stage compiler design we can easily support more language descriptions and preprocessing optimizations that when chained can greatly improve runtime efficiency in such domains. In this paper we evaluate such optimizations with the HyperTensioN HTN planner, winner of the HTN IPC 2020 total-order track.

Online goal recognition in continuous domains poses two central challenges: efficiently encoding large trajectories and effectively comparing them. Recent work addresses these challenges by using custom state-space representations and metrics to compare observations against hypotheses. However, these approaches often overlook well-established encoding techniques used in other domains that offer substantial advantages. This paper introduces a novel method for online goal recognition that leverages path signatures, a compact, expressive representation of rough path theory that captures key semantic features of trajectories in an efficient way, enabling more meaningful comparisons between them. Experiments show that our method consistently outperforms the state of the art in predictive accuracy and online planning efficiency, while remaining competitive offline.

HTN planning is a variation of classical planning where, instead of searching for a linear sequence of actions, an algorithm decomposes higher-level tasks using a method library until only executable actions remain. On one hand, this allows one to introduce domain knowledge that can speed up the search for a solution through the method library. On the other hand, it creates challenges that go beyond those of classical state-space search. While recent research produced a number of heuristics and novel algorithms that speed up HTN planning, these heuristics are not yet as informative as those available in classical planning algorithms. We investigate whether large language models (LLMs) can generate effective search heuristics for HTN planning, extending the methodology of Corrêa, Pereira, and Seipp (2025) from classical to hierarchical planning. Using the Pytrich planner on six standard total-order HTN benchmark domains, we evaluate heuristics generated by nine LLMs under domain-specific prompting and compare them against the TDG and LMCount domain-independent baselines and the PANDA planner. Our results show that LLM-generated heuristics nearly match the coverage of the best available HTN planner, while substantially reducing search effort on 83

In natural language processing (NLP), generating semantically-rich representations of sentences can improve performance on multiple tasks, such as question answering, duplicate detection, sentiment analysis, and machine translation. Recent approaches to NLP using machine learning can produce text representations that carry syntactic and semantic information. This article surveys recent works on generating sentence representations for semantic textual similarity tasks. We conduct our survey using a systematic literature review approach. We retrieve papers from several digital libraries and summarize their key techniques and findings. We propose a taxonomy to facilitate the understanding of the semantic textual similarity task on the sentence level. In our analysis, we describe the current state-of-the-art in sentence representation for semantic textual similarity and propose a guideline for working on this task.

Understanding an agent's goals from its behavior is fundamental to aligning AI systems with human intentions. Existing goal recognition methods typically rely on an optimal goal-oriented policy representation, which may differ from the actor's true behavior and hinder the accurate recognition of their goal. To address this gap, this paper introduces Goal Recognition Alignment through Imitation Learning (GRAIL), which leverages imitation learning and inverse reinforcement learning to learn one goal-directed policy for each candidate goal directly from (potentially suboptimal) demonstration trajectories. By scoring an observed partial trajectory with each learned goal-directed policy in a single forward pass, GRAIL retains the one-shot inference capability of classical goal recognition while leveraging learned policies that can capture suboptimal and systematically biased behavior. Across the evaluated domains, GRAIL increases the F1-score by more than 0.5 under systematically biased optimal behavior, achieves gains of approximately 0.1-0.3 under suboptimal behavior, and yields improvements of up to 0.4 under noisy optimal trajectories, while remaining competitive in fully optimal settings. This work contributes toward scalable and robust models for interpreting agent goals in uncertain environments.

Developing autonomous agents to deal with real-world problems is challenging, especially when developers are not necessarily specialists in artificial intelligence. Recent advances in machine learning to address natural language processing tasks are reaching performance levels suitable for practical applications, although these approaches rely on opaque and inscrutable models. This poses three key challenges: the interface of the programming with the developer, the efficiency of the resulting agents, and the scrutiny of their behaviour. Purpose: We tackle the challenge of developing autonomous agents over natural language environments in an efficient agent architecture that leverages recent developments in natural language processing, and the intuitive folk psychology abstraction of the beliefs, desires, intentions (BDI) architecture. Methods: This article introduces NatBDI, a new class of agent architecture that uses the BDI reasoning cycle with components driven by natural language processing. The resulting architecture handles natural language environments using a combination of language models and natural language inference to bootstrap the agent’s reasoning processing. Results: NatBDI agents leverage natural language components based on mental attitudes, enhancing intuitive understanding of the agent’s mental state. This allows a developer to instruct the agent more directly using a combination of controlled natural language structure and natural language knowledge as its programming interface.We empirically assess the efficiency gains of this combination while introducing a more intuitively programmed autonomous agent. Instructions in this interface substantially improve agent performance in the experimental scenario over a baseline agent created using a pure machine learning approach. Conclusion: The resulting architecture shows that combining learned policies with intuitively engineered domain knowledge yields substantial performance gains. We expect this class of agents to provide a powerful, yet intuitive, tool for agent-driven programming.

Goal Recognition (GR) aims to infer an observed agent’s intended goal frompartial observations of its behavior. Existing GR methods typically fall into two contrasting paradigms: goal-agnostic approaches, which rely on general world models and often incur heavy inference-time computation, and goal-conditioned approaches, which shift computation offline by learning goal-specific behavioral models, but struggle to generalize beyond the goals seen during training. In this paper, we introduce a unifying formulation of GR based on the type of domain knowledge required, which clarifies the trade-offs that define existing methods. Building on this formulation, we present GRUF, a new GR algorithm that leverages universal value function approximators to learn from goal-conditioned trajectories while generalizing to previously unseen goals at inference time.GRUF provides fast, lightweight online inference with a fixed-size model independent of the number of candidate goals. We evaluate GRUF across discrete andcontinuous domains, comparing it to leading GR approaches, including planning-based methods, deep RL–based recognizers, and metric-learning methods. We vary the experimental settings across levels of noise, partial observability, and the number of goals. Our results show that GRUF achieves recognition accuracycompetitive with state-of-the-art goal-conditioned methods while offering superior scalability in training time, storage, and the ability to recognize unseen goals.

Large language models have recently reached near-parity with classical planners on well-known planning domains, yet this competence relies on world-knowledge exploitation rather than genuine symbolic reasoning. Goal recognition is a complementary abductive task structurally better suited to LLM strengths: it consists of evaluating consistency with world knowledge rather than generating novel action sequences. This paper provides the first systematic zero-shot evaluation of frontier LLMs as goal recognisers on key classical PDDL benchmarks. Our results show that LLM competence on goal recognition is uneven: some models scale with evidence and approach landmark-based accuracy at full observations, while others remain anchored to world-knowledge priors regardless of how much evidence accumulates. Qualitative analysis of model reasoning traces reveals that this divergence reflects a fundamental difference in evidence integration rather than domain familiarity. These findings position goal recognition as a principled benchmark for the foundational planning knowledge of LLMs.

Deploying Large Language Model (LLM) applications, particularly those relying on Retrieval-Augmented Generation (RAG), remains challenging due to high computational demands, outdated knowledge bases, and the need to manually select optimal pipeline components. In this work, we propose a modular framework for benchmarking and guiding the efficient development of RAG applications by focusing on resource telemetry and component recommendation, suggesting the best components for a domain-specific dataset. Our approach leverages core techniques in LLM applications, including document chunking, vector databases, embedding models, and retrievers, to evaluate trade-offs among accuracy, efficiency, and scalability. By directly correlating retrieval and generation quality with underlying hardware constraints, RAGe supports researchers to identify the most effective, domain-specific RAG setups for their specific operational needs, facilitating rapid prototyping even on consumer-grade hardware.

Imitation learning stands at a crossroads: despite decades of progress, current imitation learning agents remain sophisticated memorisation machines, excelling at replay but failing when contexts shift or goals evolve. This paper argues that this failure is not technical but foundational: imitation learning has been optimised for the wrong objective. We propose a research agenda that redefines success from perfect replay to compositional adaptability. Such adaptability hinges on learning behavioural primitives once and recombining them through novel contexts without retraining. We establish metrics for compositional generalisation, propose hybrid architectures, and outline interdisciplinary research directions drawing on cognitive science and cultural evolution. Agents that embed adaptability at the core of imitation learning thus have an essential capability for operating in an open-ended world.

Multi-agent goal recognition asks an observer to jointly infer which agents act together and what each team is trying to achieve, so the hypothesis space grows combinatorially with the number of team partitions and goals per team. Real applications such as drone surveillance and collaborative robotics expose only the agents' trajectory, which forces the observer to rank team-goal hypotheses from behavior alone. Multi-Agent Goal Recognition with Branch-and-Bound (MAGR-BB) addresses this setting with a shared team- and goal-conditioned policy used as the scoring model inside a factorized branch-and-bound search. On a controlled multi-agent Blocksworld benchmark, MAGR-BB returns the same top-ranked hypothesis as exhaustive search throughout the trajectory while cutting hypothesis materialization by orders of magnitude and reducing cumulative recognition runtime substantially.

Building conversational agents to help humans in domain-specific tasks is challenging since the agent needs to understand the natural language and act over it while accessing domain expert knowledge. Modern natural language processing techniques led to an expansion of conversational agents, with recent pretrained language models achieving increasingly accurate language recognition results using ever-larger open datasets. However, the black-box nature of such pretrained language models obscures the agent's reasoning and its motivations when responding, leading to unexplained dialogues. In this work, we develop a belief-desire-intention (BDI) agent as a task-oriented dialogue system to introduce mental attitudes similar to humans describing their behavior during a dialogue. We compare the BDI model with pipeline task-oriented dialogue system architecture by leveraging existing components from dialogue systems and developing the agent's intention selection as a dialogue policy. We show that combining traditional agent modelling approaches, such as BDI, with more recent learning techniques can result in efficient and scrutable dialogue systems.

In Hierarchical Task Network (HTN) planning, landmarks are facts that must hold true, and tasks or methods that must be included in every solution. Existing landmark generation techniques for HTN planning rely on the Delete and Ordering Free (DOF) relaxation and are known to be sound but incomplete, primarily due to the limitations introduced by Task Insertion. This paper presents a new landmark generation method that builds on a previous AND/OR graph-based approach, extending it to capture additional hierarchical dependencies among tasks and methods. We prove that our approach is sound and dominates existing techniques, though it remains incomplete under the DOF relaxation. Experimental results on IPC benchmarks for totally ordered problems show that our method identifies significantly more task and method landmarks across most domains, improving coverage with minimal computational overhead.

Agent interpreters based on the Beliefs, Desires, and Intentions (BDI) model traditionally perform means-ends reasoning using plan libraries composed of reactive planning rules. However, the design of such rules often imposes a heavy knowledge engineering burden on a designer, and trades off flexibility for runtime efficiency. This use of planning rules originates from the limitations of planning technology at the time of the first BDI implementations. While these limitations have gradually been overcome by the integration of various types of planning into existing BDI theories, the corresponding interpreters remain fundamentally plan-library based. In this paper, we develop a novel BDI agent architecture driven by generalised planning as means-ends reasoning, in a radical departure from existing architectures. This architecture has two key properties. First, it more closely resembles the foundations of BDI logic and reasoning. Second, it offers substantial gains in efficiency

Goal Recognition aims to infer an agent's goal from a sequence of observations. Existing approaches often rely on manually engineered domains and discrete representations. Deep Recognition using Actor-Critic Optimization (DRACO) is a novel approach based on deep reinforcement learning that overcomes these limitations by providing two key contributions. First, it is the first goal recognition algorithm that learns a set of policy networks from unstructured data and uses them for inference. Second, DRACO introduces new metrics for assessing goal hypotheses through continuous policy representations. DRACO achieves state-of-the-art performance for goal recognition in discrete settings while not using the structured inputs used by existing approaches. Moreover, it outperforms these approaches in more challenging, continuous settings at substantially reduced costs in both computing and memory. Together, these results showcase the robustness of the new algorithm, bridging traditional goal recognition and deep reinforcement learning.

It is through affective computing that we have the integration of human feelings and computing applications. One affective computing task is Speech Emotion Recognition (SER), which identifies emotions from spoken audio. Even though emotion is a universal aspect of human experience, each culture and language has different ways to express and understand emotions. So, when designing models for SER, it is common to focus on a single language. In this work, we explore VERBO, a Brazilian Portuguese dataset for categorical emotion recognition. Our main objective is to define the best way to extract acoustic features to train a classifier for SER.We compare 18 different methods to generate audio representations, grouped by handcrafted features and audio embeddings. The best representation for VERBO is TRILL embeddings, and with an SVM classifier, we achieved 92% accuracy in VERBO. As far as we know, this was the state of the art for this dataset.

In this demonstration, we develop IntentRec4Maps, a system to recognise users' intentions in real-time interactive navigation maps. IntentRec4Maps uses the Google Maps Platform as the real-world interactive map, and a well-known approach for recognising intentions in real-time. We showcase IntentRec4Maps using two different Path-Planners and a Large Language Model (LLM).

Data pipeline frameworks provide abstractions for implementing sequences of data-intensive transformation operators, automating the deployment and execution of such transformations in a cluster. Deploying a data pipeline, however, requires computing resources to be allocated in a data center, ideally minimizing the overhead for communicating data and executing operators in the pipeline while considering each operator's execution requirements. In this paper, we model the problem of optimal data pipeline deployment as planning with action costs, where we propose heuristics aiming to minimize total execution time. Experimental results indicate that the heuristics can outperform the baseline deployment and that a heuristic based on connections outperforms other strategies.

Some imitation learning methods combine behavioural cloning with self-supervision to infer actions from state pairs. However, most rely on a large number of expert trajectories to increase generalisation and human intervention to capture key aspects of the problem, such as domain constraints. In this paper, we propose Continuous Imitation Learning from Observation (CILO), a new method augmenting imitation learning with two important features: (i) exploration, allowing for more diverse state transitions, requiring less expert trajectories and resulting in fewer training iterations; and (ii) path signatures, allowing for automatic encoding of constraints, through the creation of non-parametric representations of agents and expert trajectories. We compared CILO with a baseline and two leading imitation learning methods in five environments. It had the best overall performance of all methods in all environments, outperforming the expert in two of them.