Searcharxiv⌕ Search

arXiv subjects

Stefan Wermter

Publications and source records attributed to Stefan Wermter.

At least 19 recordsLinked to original sources

BiPOP: Partial-Order Planning for Long-Horizon Bimanual Robotic Manipulation

Bimanual robotic manipulation is versatile, yet its central challenge lies not only in executing dexterous motion with each hand, but also in coordinating the two arms for long-horizon manipulation under coupled spatial and temporal constraints. Prior work has largely focused on endowing robots with human-level manipulation skills, paying less attention to the structure of bimanual behaviour itself: many practical tasks are a coordinated composition of reusable unimanual skills and a small set of synchronous bimanual skills. This makes high-level spatio-temporal coordination, rather than skill acquisition alone, a key bottleneck for efficient bimanual execution. We propose BiPOP, a framework that couples Large Language Models (LLMs) with Temporal Partial-Order Planning for long-horizon bimanual manipulation. At its core is an explicit spatio-temporal model that partitions the workspace by arm reachability and captures inter-arm dependencies, arm assignment, synchronization requirements, and opportunities for safe concurrent execution. Building on this model, an LLM translates natural-language instructions into a formal PDDL specification, while a temporal planner enforces the constraints and schedules actions, yielding plans that are logically consistent at the symbolic level and efficient through parallel execution. On simulated long-horizon tasks of varying complexity, BiPOP outperforms open-loop planning with strong reasoning models and closed-loop ReAct-style planning in success rate, planning time, and plan efficiency, and the full pipeline runs end to end on the real bimanual robot NICOL. A failure analysis further shows that LLM-only planners mainly violate spatial and temporal coordination constraints, which motivates delegating these constraints to a symbolic planner. Code: https://github.com/Kchu/BiPOP

cs.RO↗

Reading the Table: Sim-to-Real Game State Detection for the Chef's Hat Card Game

Tabletop games offer a compelling scenario to study the interaction between artificial and human agents, combining social interaction with game state tracking and physical manipulation. To assess the game state, a robot needs to precisely detect closely related game objects, such as different card designs, requiring dedicated training data that is costly to annotate in the real world. Instead, we develop a fully automated synthetic data collection pipeline, using re-projection of known 3D coordinates to generate labels automatically and domain randomization to bridge the reality gap. This approach is used to collect a dataset of 25413 samples to finetune an object detection and a segmentation model, which are combined to assign detected cards to game-relevant zones and infer the game state. When evaluated on a real NICO humanoid robot, our pipeline achieves an F1-score of 0.906 for card-to-zone assignment, thus providing the perceptual foundation for a robot that can autonomously play Chef's Hat against human players.

cs.RO↗

DreamFormer: Dream Imitation with a Transformer World Model for Language-Conditioned Robotic Manipulation

We introduce DreamFormer, a model-based agent that acquires language-conditioned, multi-task skills by imitating expert demonstrations within the latent imagination of a learned world model. DreamFormer first learns a task-agnostic Transformer world model from unstructured play data, then acquires task-specific behaviors by optimizing an intrinsic reward that aligns agent-generated rollouts with expert demonstrations in latent space. Since the policy is trained on-policy inside imagination, it is exposed to its own errors during training, mitigating the covariate shift inherent to offline behavioral cloning. To make long-horizon imagination affordable, DreamFormer encodes a high-resolution multi-view robotic observation into a single input token, avoiding both spatial downsampling and the multi-token representations used by prior Transformer world models. On the long-horizon CALVIN benchmark, DreamFormer outperforms LUMOS, the comparable model-based agent, on single-environment evaluation (2.52 vs 2.34 average tasks completed per chain of five) while keeping imagination rollouts tractable. Against HULC, the behavior cloning baseline, it nearly doubles performance on zero-shot transfer to an unseen environment (1.30 vs 0.67), indicating that dynamics learned by the world model transfer more readily than a directly cloned policy. This is consistent with the role attributed to internal models in biological agents, where a model of environment dynamics supports behavior in situations not previously encountered.

cs.RO↗

When to Make a Decision: Task-Sufficient Active Perception for Mobile Manipulator Robots

A robotic agent operating in an unknown environment must gather sufficient task-relevant information before acting. However, determining when the available information is sufficient remains a challenge. We propose a conceptual hybrid neuro-symbolic framework for task-sufficient active perception in unknown environments. After interpreting a requested task, the agent generates an initial plan and iteratively reevaluates it as new information becomes available. At each iteration, it can acquire a more informative viewpoint, inspect or manipulate candidate objects, ask the user for clarification, or perform the task. Neural perception modules update an object-centric belief state, while a symbolic decision layer determines whether the remaining uncertainty justifies further information gathering. The framework will be implemented on a mobile manipulator comprising a humanoid torso and a mobile base, with complementary evaluation in simulation. The proposed approach aims to enable robots to determine not only how to gather additional information, but also when further perception is no longer necessary.

cs.RO↗

Towards VLA-Dreamer: Refining VLA Behavior Using World Models

Vision-Language-Action models (VLAs), while showing strong potential for robot control, require massive amounts of high-quality imitation learning data. Moreover, the absence of an explicit world model casts further doubt on their control capabilities. In this concept paper, we propose a novel architecture that addresses sample efficiency in VLAs by training a predictive world model on the embedding space of the VLA's vision encoder. We hypothesize that these embeddings are action-relevant and usable for future prediction. To this end, we propose using the suggested architecture to investigate how well these embeddings predict the future based on actions, as the inability to do so would mark a key limitation of VLA architectures: the lack of a non-lossy implicit world model to simulate real-world dynamics. The proposed architecture differs from the standard world model dynamics as the loss comes from the embedding space rather than the pixel space, similar to joint embedding predictive architectures. Furthermore, the trained world model can be utilized for short-term planning tasks by sampling VLA actions given goal images. We intend to examine the richness of vision embeddings in VLAs and reduce their high data requirements through a world model that can also generate plans during inference.

cs.RO↗

MorphIK: Morphology-Conditioned Neural Inverse Kinematics for Unknown Robots

Neural models can learn to generate various solutions to the inverse kinematics problem from data, but are usually limited to a single robot. We present MorphIK, a flow-matching model that solves inverse kinematics for revolute-joint-based kinematic chains it has never seen during training. The model uses a transformer architecture to encode the robot's morphology along with the target pose. This encoding then conditions a flow-matching head that generates poses from noise. Trained on purely synthetic data from procedurally generated robots, the model reaches a precision of about 5 cm on unseen real-world robots with 6 to 9 Degrees of Freedom. For higher precision, the model serves as an excellent Prior for further optimization algorithms, reducing error to less than 1 cm after a single step of Damped Least Squares optimization and to sub-1 mm error after 3 steps in most cases. Building on flow matching's generative capabilities to produce highly diverse outputs, our model can efficiently sample the robot's null space, providing a wide variety of configurations for the same pose. Thus, overall, MorphIK allows learning and generalizing neural inverse kinematics for a multitude of known and unknown robots.

cs.RO↗

SmoLSTM: A Compact Vision-Language-Action Model with Recurrent Memory that Persists

Vision-language-action models often predict actions from only the current observation, which can leave tasks involving object occlusion or visually identical objects ambiguous without episode history. The usual countermeasure, widening the observation window, turns the horizon into a hyperparameter and lets per-step cost grow with it. We instead capture the episode in a recurrent state. SmoLSTM couples a frozen 256M-parameter SmolVLM backbone to a matrix-memory LSTM control layer in which observation tokens and action queries are unified in a single causal stream that is never reset throughout the entire episode. Recurrent-state storage is therefore O(1) in episode length. A flow-matching action head predicts chunks of 10 end-effector pose deltas and gripper commands at each control step. Our single policy, trained jointly on 7,461 demonstrations across 140 tasks and evaluated on held-out initial states, performs best, reaching 85.1% subgoal coverage and 77.5% full-task success on LIBERO-Mem with 0.04B trainable parameters, surpassing both the benchmark's own object-centric baseline and recent memory-based approaches. Resetting the recurrent state at every control step reduces full-task success to 7.0%, showing that the trained policy relies on context carried between decisions. The same model achieves 79.6% average success on standard LIBERO.

cs.RO↗

StateVLM: A State-Aware Vision-Language Model for Robotic Affordance Reasoning

Vision-language models have demonstrated strong performance across robotic perception and instruction-following tasks. However, they still struggle with precise spatial reasoning, particularly in predicting object locations and fine-grained object states. We propose StateVLM, a vision-language model designed to learn fine-grained object representations, including object localization and grasp-relevant region prediction. We introduce a joint training objective that integrates an auxiliary regression loss (ARL) with the standard causal language modeling (CLM) objective to improve numerical reasoning and spatial understanding. To evaluate whether models can move beyond category-level grounding toward state-aware spatial understanding, we introduce an open-source benchmark, Object State Affordance Reasoning (OSAR), comprising 1,172 scenes with 7,746 individual objects and their corresponding bounding boxes. Empirical experiments on the Referring Expression Comprehension datasets demonstrate that integrating ARL improves model performance compared with CLM only. Experiments on the OSAR benchmark further demonstrate that StateVLM with ARL achieves an average improvement of 5.2% over models trained with CLM only. These results show that ARL is particularly beneficial for the complex affordance reasoning tasks that require object state understanding. The joint training objective in StateVLM demonstrates that integrating an auxiliary regression objective into VLM training improves numerical reasoning, and the OSAR benchmark provides a new testbed for understanding object state affordances in robotics.

cs.CV↗

Latent Cluster Analysis for Vision-Language-Action Models

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.

cs.RO↗

A computational model of infant sensorimotor exploration in the mobile paradigm

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.

q-bio.NC↗

When May I Help You? On The Effect of Proactivity on Group Human-Robot Collaboration

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% vs. 71.42%). The strongest differences emerge when prior experience and personality are taken into account: participants with LLM experience solve the early puzzles faster in the reactive condition, and participants with prior robot experience show modified evaluations of proactive and reactive interaction as do introverted participants. These findings demonstrate that the effects of robot initiative are simultaneously shaped by users' prior experience, personality traits and more generally by the needs of the group.

cs.RO↗

Identifying the Unknown: Prompt-Free Open Vocabulary Anomaly Recognition for Robot-Object Interaction

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% AP / 57.5% AP50 for prompt-free recognition, and 59.0% AP / 72.5% AP50 if class candidates are provided. Across additional datasets, including an in-the-wild test set with 48 unique objects, AnomNOVIC reaches up to 82.6% prompt-free detection and classification accuracy. These results significantly surpass all tested open vocabulary baselines, including YOLO-World-v2, OWLv2, and YOLOE.

cs.CV↗

Extraction and Analysis of Multimodal Concepts in Vision Language Models through Sparse Autoencoders

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

cs.CV↗

A Generalized Formalism of Auto-Regressive Decoding for Speech Processing

In speech processing, most state-of-the-art sequence prediction models rely on auto-regressive (AR) strategies to generate output sequences based on the raw predictions of the model. Despite their crucial role in the inference process, a comprehensive overview of AR strategies as a unified field is lacking, due largely to implicit and multiple definitions of next-token decoding. This context complicates the choice, comparison, and evaluation of strategies, while creating inconsistencies in the characterization of approaches as auto-regressive or not. We begin by setting explicit inclusion criteria for the field of AR search in speech processing, and derive a generalized theoretical framework to categorize and report on search strategies for neural models. We show the capabilities of this formalism in simplifying the design of benchmarks centered around the decoding process, allowing for ablation studies that are focused on search strategies.

cs.SD↗

Learning to Hear Hesitation: Continual Learning for Disfluency-Aware ASR

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.

cs.CL↗

MoDiCoL: A Modular Diagnostic Continual Learning Dataset for Robust Speech Recognition

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.

cs.CL↗

Bias Leaves a Gradient Trail: Label-Free Bias Identification via Gradient Probes on Concept Decompositions

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.

cs.CV↗

The Expert Strikes Back: Interpreting Mixture-of-Experts Language Models at Expert Level

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 $\LaTeX{}$). 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.

cs.CL↗