Gesture recognition research, unlike NLP, continues to face acute data scarcity, with progress constrained by the need for costly human recordings or image processing approaches that cannot generate authentic variability in the gestures themselves. Recent advancements in image-to-video foundation models have enabled the generation of photorealistic, semantically rich videos guided by natural language. These capabilities open up new possibilities for creating effort-free synthetic data, raising the critical question of whether video Generative AI models can augment and complement traditional human-generated gesture data. In this paper, we introduce and analyze prompt-based video generation to construct a realistic deictic gestures dataset and rigorously evaluate its effectiveness for downstream tasks. We propose a data generation pipeline that produces deictic gestures from a small number of reference samples collected from human participants, providing an accessible approach that can be leveraged both within and beyond the machine learning community. Our results demonstrate that the synthetic gestures not only align closely with real ones in terms of visual fidelity but also introduce meaningful variability and novelty that enrich the original data, further supported by superior performance of various deep models using a mixed dataset. These findings highlight that image-to-video techniques, even in their early stages, offer a powerful zero-shot approach to gesture synthesis with clear benefits for downstream tasks.

Language-conditioned robotic skills make it possible to develop the high-level reasoning of Large Language Models (LLMs) for low-level robotic control. A remaining challenge is to acquire a diverse set of fundamental skills. Existing approaches either manually decompose a complex task into primary robotic actions in a top-down fashion, or bootstrap as many combinations as possible in a bottom-up fashion to cover a wider range of task possibilities. These decompositions or combinations, however, require an initial skill library. For example, a “grasping” capability can never emerge from a skill library containing only diverse “pushing” skills. Existing skill discovery techniques with reinforcement learning acquire skills by exhaustive exploration but often yield non-meaningful behaviors. In this study, we introduce a novel learning framework for autonomous robot skill discovery that is entirely driven by LLMs. The framework begins with an LLM generating task proposals based on the provided scene description and the robot’s configurations, aiming to incrementally acquire new skills upon task completion. For each proposed task, a series of reinforcement learning processes is initiated, utilizing reward and success determination functions sampled by the LLM to develop the corresponding policy. The reliability and trustworthiness of learned behaviors are ensured by an independent vision-language model. We show that starting with zero skill, the skill library emerges and expands to more and more meaningful and reliable skills, enabling the robot to further propose and complete advanced tasks efficiently.

Mixture-of-Experts (MoE) architectures have become the dominant choice for scaling Large Language Models (LLMs), activating only a subset of parameters per token. While MoE architectures are primarily adopted for computational efficiency, it remains an open question whether their sparsity makes them inherently easier to interpret than dense feed-forward networks (FFNs). We compare MoE experts and dense FFNs using k-sparse probing and find that expert neurons are consistently less polysemantic, with the gap widening as routing becomes sparser. This suggests that sparsity pressures both individual neurons and entire experts toward monosemanticity. Leveraging this finding, we zoom out from the neuron to the expert level as a more effective unit of analysis. We validate this approach by automatically interpreting hundreds of experts. This analysis allows us to resolve the debate on specialization: experts are neither broad domain specialists (e.g., biology) nor simple token-level processors. Instead, they function as fine-grained task experts, specializing in linguistic operations or semantic tasks (e.g., closing brackets in ). Our findings suggest that MoEs are inherently interpretable at the expert level, providing a clearer path toward large-scale model interpretability. Code is available at: https://github.com/jerryy33/MoE_analysis.

Vision Language Models (VLMs) have demonstrated impressive performance in tasks requiring joint understanding of images and text, such as image captioning and Visual Question Answering (VQA), but our understanding of their internal processes remains limited. Recently, Sparse Autoencoders (SAEs) have emerged as a promising tool to support the interpretation of concepts encoded in VLMs. However, most SAE-based approaches focus only on textual or visual concepts separately, ignoring multimodal concepts. This limitation hinders a comprehensive understanding of VLMs, since concepts that integrate both modalities can be misclassified. Moreover, previous visual approaches often produce low-quality visual concept descriptions that are vague or incomplete, limiting their usefulness for understanding model reasoning. We propose a framework based on SAEs to extract and analyze visual, textual, and multimodal concepts from VLMs. For each neuron, we propose a candidate human-interpretable concept and compute the alignment between the concept and the dataset samples using cosine similarity scores. Experiments on a VQA dataset (LLaVA-NeXT) demonstrate that our framework improves visual concept quality by up to 45% compared to existing SAE-based methods, while maintaining high textual concept quality and enabling systematic identification of multimodal concepts. This work contributes new insights into the conceptual space of VLMs, providing a structured approach to distinguish between visual, textual, and multimodal concepts. The code is available at https://github.com/PHDLanza/Multidata_SAE

Large Language Models (LLMs) are increasingly used for knowledge-based reasoning tasks, yet understanding when they rely on genuine knowledge versus superficial heuristics remains challenging. We investigate this question through entity comparison tasks by asking models to compare entities along numerical attributes (e.g., “Which river is longer, the Danube or the Nile?”), which offer clear ground truth for systematic analysis. Despite having sufficient numerical knowledge to answer correctly, LLMs frequently make predictions that contradict this knowledge. We identify three heuristic biases that strongly influence model predictions: entity popularity, mention order, and semantic co-occurrence. For smaller models, a simple logistic regression using only these surface cues predicts model choices more accurately than the model's own numerical predictions, suggesting heuristics largely override principled reasoning. Crucially, we find that larger models (32B parameters) selectively rely on numerical knowledge when it is more reliable, while smaller models (7–8B parameters) show no such discrimination, which explains why larger models outperform smaller ones even when the smaller models possess more accurate knowledge. Chain-of-thought prompting steers all models towards using the numerical features across all model sizes.

In general, a similarity threshold (i.e., a vigilance parameter) for a node learning process in Adaptive Resonance Theory (ART)-based algorithms has a significant impact on clustering performance. In addition, an edge deletion threshold in a topological clustering algorithm plays an important role in adaptively generating well-separated clusters during a self-organizing process. In this paper, we propose a new parameter-free ART-based topological clustering algorithm capable of continual learning by introducing parameter estimation methods. Experimental results with synthetic and real-world datasets show that the proposed algorithm has superior clustering performance to the state-of-the-art clustering algorithms without any parameter pre-specifications.

Vision classifiers can exploit spurious correlations, achieving high in-distribution accuracy yet failing under distribution shift. Existing approaches to bias mitigation and analysis often depend on curated datasets, spurious-attribute or group labels, or retraining, which may be infeasible once a model is deployed or the relevant bias is unknown. We present a bias-label-free, post-hoc method for identifying spurious concepts in frozen vision models, relying only on standard class labels from a held-out audit dataset. For each target class, we collect patches from inputs predicted as that class and apply non-negative matrix factorization to intermediate activations to obtain a bank of interpretable concept vectors. Candidate concepts are then ranked with a bias estimator derived from their interaction with backpropagated gradients on misclassified examples: bias concepts tend to get activated when correcting false negatives and suppressed when correcting false positives. On Colored MNIST and Waterbirds the method recovers concepts aligned with the known spurious cue, and on CelebA it surfaces decision-relevant directions that only partially coincide with the annotated gender attribute; suppressing the top-ranked concepts at inference time improves worst-group accuracy by up to 17.9 percentage points on Waterbirds and 10.4 on CelebA without any retraining or parameter updates. Our method identifies decision-relevant spurious directions that need not coincide with annotated ones, providing both an interpretable auditing tool and an actionable debiasing handle for frozen vision models. Code is available at https://github.com/vitryt/label-free-bias-identification.

Robot initiative is a central challenge in multi-party human-robot collaboration. A robot that contributes without being addressed may provide timely support, but it may also disrupt coordination, divide attention, or interrupt turn-taking; a robot that waits to be addressed may preserve human control, but it may also miss opportunities to assist. We investigate this design challenge in a collaborative escape room in which pairs of participants work with a humanoid robot under either a reactive interaction model, where the robot responds only when addressed, or a proactive model, where it listens continuously, contributes autonomously, and periodically re-initiates interaction. We evaluate both models using puzzle-solving performance, interaction frequency, and participant ratings on the Godspeed and RoSAS scales. The proactive model substantially increases interaction frequency, whereas the reactive model shows a descriptively higher overall success rate (92.86

Video recordings of interviews, lectures, and meetings contain valuable moments surrounded by less essential talk. Making a shareable and meaningful shorter version of this content requires significant effort because it combines tedious, repeated operations with personal editorial decisions, which require human judgment. We introduce an editing approach that operates on video transcripts and combines a three-stage large language model pipeline with a timeline-anchored, marker-based interface so editors can inspect and refine suggestions before final assembly. The pipeline first produces an overview summary to maximize content coverage, then induces plain-language selection rules that encode editorial intent, and finally applies rule-conditioned ranking on small transcript windows to mitigate long-context limits, yielding strictly extractive, time-aligned spans under duration constraints. The interface displays groupings of short excerpts using markers with priorities and confidence cues, converting opaque model output into verifiable units within standard video editing workflows. On MeetingBank and MeetingBank-QA datasets, our method outperforms practical extractive baselines at matched lengths. In a within-subjects study with experienced video editors familiar with Premiere Pro video editing software, we found that our marker-based interface provided editors higher efficiency, control, and satisfaction than both a manual editing baseline and an opaque auto-cut condition.

Reward models trained through Reinforcement Learning from AI Feedback (RLAIF) methods frequently suffer from limited generalizability, which hinders the alignment performance of policy models. This challenge stems from various issues, including distribution shift, preference label noise, and mismatch of overly challenging samples with model capacity. In this paper, we aim to enhance the generalizability of reward models through a data-centric approach, driven by the insight that these issues are inherently intertwined from a uniform perspective of data difficulty. Accordingly, we propose a novel framework, Curriculum-RLAIF, which constructs preference pairs with varying difficulty levels and then produces a specific curriculum for reward model training. Comprehensive experimental results suggest that reward models trained with Curriculum-RLAIF achieve improved generalizability, boosting the alignment performance of policy models by a significant margin without incurring additional inference costs compared to various existing non-curriculum baselines. Further analysis and comparison with alternative strategies highlight the superiority of Curriculum-RLAIF in simplicity, efficiency, and effectiveness.

Uncertainty, vagueness, and ambiguity are closely related and often confused concepts in human-robot interaction (HRI). In earlier studies, these concepts have been defined in contradictory ways and described using inconsistent terminology. This conceptual confusion and lack of terminological consistency undermine empirical comparability, thereby slowing the accumulation of theory. Consequently, consistent concepts that clarify these challenges, including their definitions, distinctions, and interrelationships, are needed in HRI. To address this lack of clarity, this paper proposes a consistent conceptual foundation for the challenges of uncertainty, vagueness, and ambiguity in HRI. First, we examine the meanings of these three terms in dictionaries. We then analyze the nature of their distinctions and interrelationships within the context of HRI. We further illustrate these characteristics through examples. Finally, we demonstrate how this consistent conceptual foundation facilitates the design of novel methods and the evaluation of existing methodologies for these phenomena.

In conversations, human emotions are transient; however, they tend to persist across multiple utterances. For example, we rarely switch instantly between contrasting emotions such as happiness and anger. Instead, emotions tend to evolve smoothly, and these patterns are often speaker-specific. Some people might escalate, while others gradually cool down over time. Furthermore, when emotions change during a conversation, they are often driven by contextual factors, such as newly received information or unexpected events. Even though progress has been made in Emotion Recognition in Conversations (ERC), most existing approaches still rely heavily on overt evidence and do not sufficiently model these non-apparent factors. Especially in multimodal settings, this makes these models fragile when the signals are noisy (e.g., occluded faces, slang expressions, or microphone noise). To address these limitations, we introduce Speaker-Conditioned Priors over Emotions (SCoPE). SCoPE is a light weight module that utilizes the emotional history of each speaker and explicitly models their priors for use in subsequent emotion classification. Second, we incorporate emotion shift prediction, a well-established concept in ERC, to guide the model in balancing the priors from SCoPE and multimodal evidence. Finally, we propose a shift-aware fusion mechanism that performs precision-weighted logit integration between multimodal evidence and the speaker prior, forming a Bayesian-inspired product-of-experts formulation. This dynamic fusion allows the model to rely on historical priors when emotions persist and to prioritize multimodal evidence when shifts are likely. Experimental results show our model achieves superior performance over recent state-of-the-art models on the IEMOCAP dataset in multimodal settings.

Vision-language models (VLMs) have shown remarkable performance in various robotic tasks, as they can perceive visual information and understand natural language instructions. However, when applied to robotics, VLMs remain subject to a fundamental limitation inherent in large language models (LLMs): they struggle with numerical reasoning, particularly in object detection and object-state localization. To explore numerical reasoning as a regression task in VLMs, we propose a novel training strategy to adapt VLMs for object detection and object-state localization. This approach leverages box decoder outputs to compute an Auxiliary Regression Loss (ARL) during fine-tuning, while preserving standard sequence prediction at inference. We leverage this training strategy to develop StateVLM (State-aware Vision-Language Model), a novel model designed to perceive and learn fine-grained object representations, including precise localization of objects and their states, as well as graspable regions. Due to the lack of a benchmark for object-state affordance reasoning, we introduce an open-source benchmark, Object State Affordance Reasoning (OSAR), which contains 1,172 scenes with 7,746 individual objects and corresponding bounding boxes. Comparative experiments on adapted benchmarks (RefCOCO, RefCOCO+, and ) demonstrate that ARL improves model performance by an average of 1.6% compared to models without ARL. Experiments on the OSAR benchmark further support this finding, showing that StateVLM with ARL achieves an average of 5.2% higher performance than models without ARL. In particular, ARL is also important for the complex task of affordance reasoning in OSAR, where it enhances the consistency of model outputs.

Robots operating in real-world environments must in general be able to recognize previously unseen objects. As robotic systems move toward open-world autonomy, there is a growing, yet largely unmet, need for open vocabulary object detectors that are prompt-free and efficient enough for continuous deployment. We present AnomNOVIC, a two-stage known-workspace framework that combines a masked autoencoder (MAE) trained for anomaly detection, with NOVIC, a powerful real-time prompt-free open vocabulary image classifier. The MAE produces generic object-agnostic bounding boxes, allowing NOVIC to classify salient image regions without requiring a predefined candidate class list. We evaluate AnomNOVIC against strong open vocabulary baselines in a tabletop robot-object environment featuring the NICOL humanoid robot, reaching 47.1

Modern Automatic Speech Recognition (ASR) systems have made remarkable progress on standard benchmarks, yet performance gaps have emerged under real-world distribution shifts, caused by recording conditions, accents, speech impairments, and noise. Existing datasets and benchmarks typically isolate these factors, which overlooks their co-occurrence in real-world applications. In this paper, we argue that model robustness can be treated as a dynamic capability that continually develops, and we introduce MoDiCoL, a Modular Diagnostic Continual Learning dataset designed for controlled analysis of linguistic content, speaker characteristics, and acoustic environments. Furthermore, we propose a real-world-inspired continual learning curriculum to simulate incremental updates and study how robustness is acquired, transferred, and forgotten. We evaluate three continual learning strategies and provide detailed insights into robustness under evolving conditions.

Despite advances in large-scale Automatic Speech Recognition (ASR), disfluent speech remains challenging, as state-of-the-art systems are often optimized to omit disfluencies, leading to information loss and hallucinations. Prior work has focused on verbatim transcription and the integration of disfluency markers, but adapting models on limited datasets can lead to catastrophic forgetting of general-domain knowledge. We address this gap by leveraging continual learning (CL) with explicit disfluency tokens. We first introduce these tokens into a pretrained ASR model to establish stable token mechanisms, and then continue training on additional datasets with varying disfluency distributions. Through a detailed analysis of model dynamics during training, we identify a trade-off between marker learning and ASR performance, and a consistent cross-attention head mechanism shared across CL methods.

We present a computational model of the mechanisms that may determine infant behavior in the "mobile paradigm." This paradigm has been used in developmental psychology to explore how infants learn the sensory effects of their actions. In this paradigm, a mobile (an articulated and movable object hanging above an infant's crib) is connected to one of the infant's limbs, prompting the infant to preferentially move that "connected" limb. This ability to detect a "sensorimotor contingency" is considered to be a foundational cognitive ability in development. To understand how infants learn sensorimotor contingencies, we built a model that attempts to replicate infant behavior. Our model incorporates a neural network, action-outcome prediction, exploration, motor noise, preferred activity level, and biologically inspired motor control. We find that simulations with our model replicate the classic findings in the literature showing preferential movement of the connected limb. An interesting observation is that the model sometimes exhibits a burst of movement after the mobile is disconnected, shedding light on a similar occasional finding in infants. In addition to these general findings, the simulations also replicate data from two recent more detailed studies using a connection with the mobile that was either gradual or all-or-none. A series of ablation studies further shows that the inclusion of mechanisms of action-outcome prediction, exploration, motor noise, and biologically inspired motor control was essential for the model to correctly replicate infant behavior. This suggests that these components are also involved in infant sensorimotor learning.

Learning sensorimotor contingencies—that is, the link between one’s actions and their sensory effects—is fundamental to developing body knowledge, understanding causality, and developing a sense of agency. In developmental psychology, this process is classically studied using the mobile paradigm, where infants learn that movement of a limb causes motion of a connected mobile. To expand our understanding of how infants learn this, we tested an embodied computational model that learns through two biologically inspired mechanisms: prediction and curiosity. Implemented on the child-sized iCub humanoid robot interacting with a mobile, the model detected sensorimotor contingencies across several experimental conditions using a variety of movement strategies. Our findings suggest that contingency learning cannot be captured by a single behavioral metric, such as the amount of movement, but instead emerges through a spectrum of exploratory behaviors. Analysis of the robot’s internal activity reveals that these behaviors emerge from the dynamic trade-off between prediction and curiosity—between exploitation and exploration. Our work provides a biologically motivated, physically embodied model of sensorimotor interaction that connects theories of infant learning with robotic implementations. The results allow us to generate testable hypotheses for developmental research and to inform the design of autonomous learning systems.

We introduce a Generalizable Neural Radiance Field approach for predicting 3D workspace occupancy from egocentric robot observations. Unlike prior methods operating in camera-centric coordinates, our model constructs occupancy representations in a global workspace frame, making it directly applicable to robotic manipulation. The model integrates flexible source views and generalizes to unseen object arrangements without scene-specific finetuning. We demonstrate the approach on a humanoid robot and evaluate predicted geometry against 3D sensor ground truth. Trained on 40 real scenes, our model achieves 26mm reconstruction error, including occluded regions, validating its ability to infer complete 3D occupancy beyond traditional stereo vision methods.

Vision-Language-Action (VLA) Models are increasingly used in robotics for their ability to ground language and perception into action, yet the internal representations driving their behaviour remain poorly understood. We propose LAVLA, a framework for latent cluster analysis of VLA models, and conduct a layer-wise study of the state-of-the-art GR00T N1.5 model, with particular focus on its action decoder. To better characterise the latent space during action diffusion, we introduce a cross-attention-based embedding-weighting method that amplifies relevant features while suppressing less informative ones. Quantitative evaluation shows that weighted clustering consistently outperforms the baseline. To improve interpretability, we extract human-interpretable concepts for each cluster, linking latent representations to semantic descriptions. Our analysis shows that latent clusters progressively disentangle spatiotemporal and kinematic features, with representations becoming more refined in the middle layers and stabilising toward the output. As such, LAVLA advances the interpretability of language-driven robotic systems.