Search arXiv⌕ Search

arXiv subjects

Elmar Rueckert

Publications and source records attributed to Elmar Rueckert.

At least 19 recordsLinked to original sources
arXiv2026

InfiNoVA: Infinite Novel View Augmentation for Viewpoint Invariant Robot Policies

Vision-Language-Action (VLA) policies often rely strongly on the camera viewpoints seen during training, causing substantial performance degradation when deployed from unseen perspectives. Collecting demonstrations from sufficiently diverse physical viewpoints is expensive and still provides only sparse coverage of the viewpoint space. We introduce InfiNoVA, a data-augmentation framework that converts synchronized multi-camera demonstrations into a dense distribution of geometrically consistent training views. InfiNoVA reconstructs each manipulation trajectory as a time-varying 3D Gaussian representation and renders novel observations from sampled camera poses while preserving the original state-action correspondence. This explicit scene representation improves frame-level fidelity and temporal consistency while reducing task-critical hallucinations observed in generative novel-view synthesis. Across four real-world manipulation tasks, policies trained with InfiNoVA achieve 5.4x higher average success under unseen randomized viewpoints than both VISTA-based augmentation and the unaugmented policy. InfiNoVA further achieves 1.7x higher success than training directly on all five physical camera views. These results show that dense, geometrically grounded viewpoint augmentation provides a practical route toward camera-robust robot policies without modifying the underlying policy architecture.

cs.RO↗
arXiv2026

Pro-Bench: Prompt-Robust Open-Vocabulary Visual Grounding Across Real-World Heterogeneous Environments

Open-vocabulary visual grounding enables robots to localise task-relevant entities from natural-language queries without dependence on predefined perceptual taxonomies. However, existing benchmarks largely rely on short category labels and web-scraped imagery, leaving it unclear whether open-vocabulary models can robustly ground diverse queries and visual conditions under real deployments. We introduce \textbf{Pro-Bench}, a prompt-conditioned benchmark for open-vocabulary visual grounding in heterogeneous, real-world environments. Pro-Bench includes $13k+$ RGB frames from independent robotic domains (subterranean, industrial, indoor, outdoor, urban), with $74.5k$ manual instance annotations and $515$ target queries covering categorical, attributive, relational, affordance, state, part-whole, negative, and compositional semantics. We benchmarked $16$ open-vocabulary model configurations in strict zero-shot inference, measuring localisation accuracy across IoU thresholds, end-to-end inference latency, prompt-induced performance variation, and target recovery consistency. Our results show that prompt-robustness is strongly architecture-dependent. Most model configurations ($10/16$) perform best with short category labels, whereas free-form queries yield the highest accuracy for only one. Moreover, similar aggregate mAP can conceal substantial differences in consistent target recovery across reformulations. Pro-Bench enables systematic evaluation of these gaps and supports prompt-robust visual grounding. Pro-Bench: https://pro-bench.github.io/.

cs.CV↗
arXiv2026

Sandwich-Residuals: Parameter-Efficient Test-time Adaptation of World Models

Latent world models enable planning by predicting the effects of actions in a learned representation space, but their predictions can become unreliable when test-time conditions differ from training. Existing test-time adaptation methods address this by updating parts of the pretrained model, often modifying millions of parameters and requiring a choice of which internal components to adapt. We introduce Sandwich-Residuals, a lightweight alternative that keeps the pretrained world model frozen and learns only small residual corrections around the predictor. The residuals are optimized online using the model's self-supervised prediction error and require no rewards, labels, or source-domain data. Across 21 conditions on the AdaJEPA benchmark, our method achieves $1.3\times$ the success rate of the frozen model while retaining 95% of the performance of the strongest AdaJEPA variant and adapting 97-99% fewer parameters. Under compound shifts, this advantage increases to $1.9\times$ the success rate of the frozen model, while remaining comparable to internal block adaptation. We further demonstrate the same adaptation principle on a DINO-WM model for 3-D manipulation. These results suggest that effective test-time adaptation of world models does not necessarily require modifying their pretrained internal weights.

cs.RO↗
arXiv2026

EliGSiR: Continual RGB-D Mapping with Gaussian Splatting under Bounded Compute

Conventional 3D Gaussian Splatting assumes a closed set of observations and long optimization schedules. Continual RGB-D mapping in contrast poses the problem that new observations arrive online, while previously reconstructed regions must be preserved. We present EliGSiR (Evidence-guided Load-adaptive Incremental Gaussian Splatting with Image Replay), a continual Gaussian mapper that controls how the available optimization budget is used as the reconstruction evolves. Map-Guided View Scheduling filters redundant incoming views and reconsiders retained views according to the current state of the map. Load-Adaptive Fidelity adjusts supervision resolution to the current mapping load instead of following a fixed resolution schedule. Targeted Geometry Growth separates depth supervision from Gaussian creation and adds geometric capacity only where repeated RGB-D observations indicate missing or misplaced structure. Together, these mechanisms adapt which views are optimized, how much image detail is used, and where the representation grows while mapping remains active. We evaluate EliGSiR on Replica, TUM RGB-D, ScanNet++, and real RGB-D sensor sequences, considering both the final reconstruction and the map available throughout acquisition. On TUM RGB-D fr3/long_office_household, EliGSiR reaches 21.52 dB with the same ground-truth mapping poses used by the controlled baselines, compared with 19.42 dB for SplaTAM. In the tracked-pose comparison, EliGSiR with live ORB-SLAM3 poses reaches 23.02 dB in 155.5 s, compared with 20.10 dB in 230.9 s for CaRtGS using its native tracker. We further evaluate reconstruction throughout acquisition and show how EliGSiR adaptive view scheduling, supervision fidelity, and geometry growth improve the use of the available mapping budget.

cs.CV↗
arXiv2026

VLEM: Real-Time 3D Vision-Language Embedding Mapping

Semantic scene understanding in robotics requires representations that are both metric-accurate and queryable via natural language in real-time. While recent Vision-Language Models enable powerful 2D image-text alignment, their integration into real-time 3D mapping systems remains challenging due to their requirements on ground truth poses, computational cost, and memory constraints. We present VLEM (Vision-Language Embedding Mapping), a real-time framework for integrating pixel-aligned 2D vision-language embeddings from various backends into a globally consistent, metric-accurate 3D representation, requiring only a raw RGB-D stream. Compared to ConceptFusion, Open-Fusion, and RayFronts, VLEM provides better open-set segmentation performance and a more compact representation. We further demonstrate VLEM's versatility in interactive real-time robotic manipulation tasks and mobile mapping scenarios.

cs.RO↗
arXiv2026

ReLI: Cross-Lingual Language-to-Action Grounding for Human-Robot Interaction

Adapting autonomous agents for real-world industrial, domestic, and other daily tasks is currently gaining momentum. However, in global or cross-lingual application contexts, the ability to instruct these agents in one's native language remains until today a formidable challenge. Existing language-conditioned human-robot interaction frameworks typically support only a handful of high-resource languages, e.g., English and Chinese, limiting accessibility for billions of potential end users. To address this gap, we propose ReLI, a cross-lingual framework that enables autonomous agents to converse naturally, reason semantically about their environment, and execute downstream tasks, regardless of the tasks' instruction linguistic origin or input modalities. We ground large-scale pre-trained foundation models and transform them into language-to-action models that can directly provide common-sense reasoning and high-level robot control through free-form conversational interactions. We then perform an implicit language-conditioned cross-lingual adaptation of the models to ensure that ReLI generalises effectively across diverse global languages. We conducted extensive empirical evaluation on a diverse set of short- and long-horizon tasks, including zero-shot and few-shot spatial navigation, scene information retrieval, and query-oriented tasks, and then benchmarked the performance across more than $70K+$ multi-turn conversations in over $140$ languages spanning high-resource, low-resource, and vulnerable/creole tiers. Across the benchmarked languages, ReLI achieved consistently high instruction-parsing accuracy, task success rate, and rapid response time. Further, we c..

cs.RO↗
arXiv2026

SIL: Symbiotic Interactive Learning for Language-Conditioned Human-Agent Co-Adaptation

Today's autonomous agents, largely driven by foundation models (FMs), can understand natural language instructions and solve long-horizon tasks with human-like reasoning. However, current human-robot interaction frameworks largely follow a one-way master-apprentice technique where the embodied agent passively executes commands without reciprocal learning. This neglects the co-adaptive, multi-turn nature of everyday human-to-human interactions. We introduce symbiotic interactive learning (SIL), a bidirectional co-adaptation framework in a shared latent task space, where both the human and the agent maintain joint belief states that evolve with the interaction history. This enables proactive clarification, adaptive suggestions, and shared plan refinement. SIL leverages FMs for spatial perception and reasoning, together with a triplet-loss-trained neural encoder that grounds the FMs' outputs into task-specific latent representations. To support long-term stability as tasks evolve, SIL utilises episodic and semantic memory architectures, regularised via elastic weight consolidation to mitigate catastrophic forgetting. We evaluate SIL on simulated and real-world embodied tasks, including instruction following, information retrieval, query-oriented reasoning, and interactive dialogue, achieving a $90.4\%$ task completion rate and a belief alignment score of $ρ\approx 0.83$, an absolute improvement of about $20$ percentage points over the best ablations. Demos and resources: https://linusnep.github.io/SIL/.

cs.RO↗
arXiv2026

SteelDS: A High-Resolution Video Dataset of E40 Steel Scrap for Object Detection and Instance Segmentation

This dataset provides high-resolution, annotated video sequences of shredded E40-grade steel and copper scrap on a conveyor belt. Captured in a controlled laboratory environment, the data reflects the industrial post-magnetic sorting stage, where manual intervention is typically required to remove copper contaminants. The dataset comprises 24,297 labeled frames across five subsets, featuring 396 steel and 101 copper objects categorized by size. It supports the development of machine learning models for material classification, object detection, and instance segmentation. Variations in object spacing and density are included to simulate realistic industrial sorting conditions. Ground truth annotations include pixel-wise segmentation masks and material classes. This dataset serves as a benchmark for evaluating automated sorting algorithms aiming to identify copper impurities within complex, heterogeneous steel scrap streams.

cs.RO↗
arXiv2026

PASTA: Vision Transformer Patch Aggregation for Weakly Supervised Target and Anomaly Segmentation

Detecting unseen anomalies in unstructured environments presents a critical challenge for industrial and agricultural applications such as material recycling and weeding. Existing perception systems frequently fail to satisfy the strict operational requirements of these domains, specifically real-time processing, pixel-level segmentation precision, and robust accuracy, due to their reliance on exhaustively annotated datasets. To address these limitations, we propose a weakly supervised pipeline for object segmentation and classification using weak image-level supervision called 'Patch Aggregation for Segmentation of Targets and Anomalies' (PASTA). By comparing an observed scene with a nominal reference, PASTA identifies Target and Anomaly objects through distribution analysis in self-supervised Vision Transformer (ViT) feature spaces. Our pipeline utilizes semantic text-prompts via the Segment Anything Model 3 to guide zero-shot object segmentation. Evaluations on a custom steel scrap recycling dataset and a plant dataset demonstrate a 75.8% training time reduction of our approach to domain-specific baselines. While being domain-agnostic, our method achieves superior Target (up to 88.3% IoU) and Anomaly (up to 63.5% IoU) segmentation performance in the industrial and agricultural domain.

cs.CV↗
arXiv2025

Rock Classification through Knowledge-Enhanced Deep Learning: A Hybrid Mineral-Based Approach

Automated rock classification from mineral composition presents a significant challenge in geological applications, with critical implications for material recycling, resource management, and industrial processing. While existing methods using One dimensional Convolutional Neural Network (1D-CNN) excel at mineral identification through Raman spectroscopy, the crucial step of determining rock types from mineral assemblages remains unsolved, particularly because the same minerals can form different rock types depending on their proportions and formation conditions. This study presents a novel knowledge-enhanced deep learning approach that integrates geological domain expertise with spectral analysis. The performance of five machine learning methods were evaluated out of which the 1D-CNN and its uncertainty-aware variant demonstrated excellent mineral classification performance (98.37+-0.006% and 97.75+-0.010% respectively). The integrated system's evaluation on rock samples revealed variable performance across lithologies, with optimal results for limestone classification but reduced accuracy for rocks sharing similar mineral assemblages. These findings not only show critical challenges in automated geological classification systems but also provide a methodological framework for advancing material characterization and sorting technologies.

cs.CE↗
arXiv2025

Privacy-Aware Lifelong Learning

Lifelong learning algorithms enable models to incrementally acquire new knowledge without forgetting previously learned information. Contrarily, the field of machine unlearning focuses on explicitly forgetting certain previous knowledge from pretrained models when requested, in order to comply with data privacy regulations on the right-to-be-forgotten. Enabling efficient lifelong learning with the capability to selectively unlearn sensitive information from models presents a critical and largely unaddressed challenge with contradicting objectives. We address this problem from the perspective of simultaneously preventing catastrophic forgetting and allowing forward knowledge transfer during task-incremental learning, while ensuring exact task unlearning and minimizing memory requirements, based on a single neural network model to be adapted. Our proposed solution, privacy-aware lifelong learning (PALL), involves optimization of task-specific sparse subnetworks with parameter sharing within a single architecture. We additionally utilize an episodic memory rehearsal mechanism to facilitate exact unlearning without performance degradations. We empirically demonstrate the scalability of PALL across various architectures in image classification, and provide a state-of-the-art solution that uniquely integrates lifelong learning and privacy-aware unlearning mechanisms for responsible AI applications.

cs.LG↗
arXiv2025

Amending CALPHAD databases using a neural network for predicting mixing enthalpy of liquids

In order to establish the thermodynamic stability of a system, knowledge of its Gibbs free energy is essential. Most often, the Gibbs free energy is predicted within the CALPHAD framework using models employing thermodynamic properties, such as the mixing enthalpy, heat capacity, and activity coefficients. Here, we present a deep-learning approach capable of predicting the mixing enthalpy of liquid phases of binary systems that were not present in the training dataset. Therefore, our model allows for a system-informed enhancement of the thermodynamic description to unknown binary systems based on information present in the available thermodynamic assessment. Thereby, significant experimental efforts in assessing new systems can be spared. We use an open database for steels containing 91 binary systems to generate our initial training (and validation) and amend it with several direct experimental reports. The model is thoroughly tested using different strategies, including a test of its predictive capabilities. The model shows excellent predictive capabilities outside of the training dataset as soon as some data containing species of the predicted system is included in the training dataset. The estimated uncertainty of the model is below 1 kJ/mol for the predicted mixing enthalpy. Subsequently, we used our model to predict the enthalpy of mixing of all binary systems not present in the original database and extracted the Redlich-Kister parameters, which can be readily reintegrated into the thermodynamic database file.

physics.chem-ph↗
arXiv2025

EnvoDat: A Large-Scale Multisensory Dataset for Robotic Spatial Awareness and Semantic Reasoning in Heterogeneous Environments

To ensure the efficiency of robot autonomy under diverse real-world conditions, a high-quality heterogeneous dataset is essential to benchmark the operating algorithms' performance and robustness. Current benchmarks predominantly focus on urban terrains, specifically for on-road autonomous driving, leaving multi-degraded, densely vegetated, dynamic and feature-sparse environments, such as underground tunnels, natural fields, and modern indoor spaces underrepresented. To fill this gap, we introduce EnvoDat, a large-scale, multi-modal dataset collected in diverse environments and conditions, including high illumination, fog, rain, and zero visibility at different times of the day. Overall, EnvoDat contains 26 sequences from 13 scenes, 10 sensing modalities, over 1.9TB of data, and over 89K fine-grained polygon-based annotations for more than 82 object and terrain classes. We post-processed EnvoDat in different formats that support benchmarking SLAM and supervised learning algorithms, and fine-tuning multimodal vision models. With EnvoDat, we contribute to environment-resilient robotic autonomy in areas where the conditions are extremely challenging. The datasets and other relevant resources can be accessed through https://linusnep.github.io/EnvoDat/.

cs.RO↗
arXiv2024

Multimodal Human-Autonomous Agents Interaction Using Pre-Trained Language and Visual Foundation Models

In this paper, we extended the method proposed in [21] to enable humans to interact naturally with autonomous agents through vocal and textual conversations. Our extended method exploits the inherent capabilities of pre-trained large language models (LLMs), multimodal visual language models (VLMs), and speech recognition (SR) models to decode the high-level natural language conversations and semantic understanding of the robot's task environment, and abstract them to the robot's actionable commands or queries. We performed a quantitative evaluation of our framework's natural vocal conversation understanding with participants from different racial backgrounds and English language accents. The participants interacted with the robot using both spoken and textual instructional commands. Based on the logged interaction data, our framework achieved 87.55% vocal commands decoding accuracy, 86.27% commands execution success, and an average latency of 0.89 seconds from receiving the participants' vocal chat commands to initiating the robot's actual physical action. The video demonstrations of this paper can be found at https://linusnep.github.io/MTCC-IRoNL/.

cs.RO↗
arXiv2024

ED-VAE: Entropy Decomposition of ELBO in Variational Autoencoders

Traditional Variational Autoencoders (VAEs) are constrained by the limitations of the Evidence Lower Bound (ELBO) formulation, particularly when utilizing simplistic, non-analytic, or unknown prior distributions. These limitations inhibit the VAE's ability to generate high-quality samples and provide clear, interpretable latent representations. This work introduces the Entropy Decomposed Variational Autoencoder (ED-VAE), a novel re-formulation of the ELBO that explicitly includes entropy and cross-entropy components. This reformulation significantly enhances model flexibility, allowing for the integration of complex and non-standard priors. By providing more detailed control over the encoding and regularization of latent spaces, ED-VAE not only improves interpretability but also effectively captures the complex interactions between latent variables and observed data, thus leading to better generative performance.

cs.LG↗
arXiv2024

M2CURL: Sample-Efficient Multimodal Reinforcement Learning via Self-Supervised Representation Learning for Robotic Manipulation

One of the most critical aspects of multimodal Reinforcement Learning (RL) is the effective integration of different observation modalities. Having robust and accurate representations derived from these modalities is key to enhancing the robustness and sample efficiency of RL algorithms. However, learning representations in RL settings for visuotactile data poses significant challenges, particularly due to the high dimensionality of the data and the complexity involved in correlating visual and tactile inputs with the dynamic environment and task objectives. To address these challenges, we propose Multimodal Contrastive Unsupervised Reinforcement Learning (M2CURL). Our approach employs a novel multimodal self-supervised learning technique that learns efficient representations and contributes to faster convergence of RL algorithms. Our method is agnostic to the RL algorithm, thus enabling its integration with any available RL algorithm. We evaluate M2CURL on the Tactile Gym 2 simulator and we show that it significantly enhances the learning efficiency in different manipulation tasks. This is evidenced by faster convergence rates and higher cumulative rewards per episode, compared to standard RL algorithms without our representation learning approach.

cs.RO↗
arXiv2024

Integrating Human Expertise in Continuous Spaces: A Novel Interactive Bayesian Optimization Framework with Preference Expected Improvement

Interactive Machine Learning (IML) seeks to integrate human expertise into machine learning processes. However, most existing algorithms cannot be applied to Realworld Scenarios because their state spaces and/or action spaces are limited to discrete values. Furthermore, the interaction of all existing methods is restricted to deciding between multiple proposals. We therefore propose a novel framework based on Bayesian Optimization (BO). Interactive Bayesian Optimization (IBO) enables collaboration between machine learning algorithms and humans. This framework captures user preferences and provides an interface for users to shape the strategy by hand. Additionally, we've incorporated a new acquisition function, Preference Expected Improvement (PEI), to refine the system's efficiency using a probabilistic model of the user preferences. Our approach is geared towards ensuring that machines can benefit from human expertise, aiming for a more aligned and effective learning process. In the course of this work, we applied our method to simulations and in a real world task using a Franka Panda robot to show human-robot collaboration.

cs.RO↗
arXiv2024

Multimodal Visual-Tactile Representation Learning through Self-Supervised Contrastive Pre-Training

The rapidly evolving field of robotics necessitates methods that can facilitate the fusion of multiple modalities. Specifically, when it comes to interacting with tangible objects, effectively combining visual and tactile sensory data is key to understanding and navigating the complex dynamics of the physical world, enabling a more nuanced and adaptable response to changing environments. Nevertheless, much of the earlier work in merging these two sensory modalities has relied on supervised methods utilizing datasets labeled by humans.This paper introduces MViTac, a novel methodology that leverages contrastive learning to integrate vision and touch sensations in a self-supervised fashion. By availing both sensory inputs, MViTac leverages intra and inter-modality losses for learning representations, resulting in enhanced material property classification and more adept grasping prediction. Through a series of experiments, we showcase the effectiveness of our method and its superiority over existing state-of-the-art self-supervised and supervised techniques. In evaluating our methodology, we focus on two distinct tasks: material classification and grasping success prediction. Our results indicate that MViTac facilitates the development of improved modality encoders, yielding more robust representations as evidenced by linear probing assessments.

cs.RO↗