arXiv subjects

Sven Behnke

Publications and source records attributed to Sven Behnke.

At least 19 recordsLinked to original sources

arXiv2026

SlotDiT: Object-Centric Representations for Diffusion Transformers

Text-conditioned latent diffusion models perform strongly in video generation and are promising backbones for robotic applications. However, existing approaches rely on pixel-level or VAE-based latent representations that lack explicit semantic structure, leaving the impact of the representation space largely unexplored. Slot-based object-centric representations offer a structured alternative by decomposing scenes into object-level latents, or slots. While they have shown success in dynamics modeling and planning, they have not yet been explored for diffusion-based generative modeling. We introduce SlotDiT, a text-guided Diffusion Transformer (DiT) that operates in a slot-based latent space. Given a reference image and a language instruction, SlotDiT decomposes the scene into object-centric slots representing individual entities. Conditioned on the instruction and observed scene context, the model autoregressively denoises future slot trajectories to predict scene dynamics. To systematically investigate latent-space design for diffusion transformers, we compare slot-based representations against VAE-based and semantics-aligned alternatives within a unified DiT framework. Our experiments show that using slots as DiT latents yields competitive video generation quality while consistently improving task-completion rates across four robotic datasets. Furthermore, their compact representation provides a computationally efficient alternative to VAE-based and semantics-aligned latent spaces. Overall, our results demonstrate that object-centric structure is a powerful inductive bias for diffusion-based generative modeling in robotic environments. The project page is available at https://slot-dit.github.io/.

cs.CV↗

arXiv2026

StrAD: A Streaming Method and Benchmark for Audio Description Generation for Long-form Videos

Visual content is the dominant medium of communication, yet without audio descriptions (ADs), it remains inaccessible to blind and low-vision people. ADs narrate context-relevant visual events during natural audio pauses. Manually creating ADs is expensive, limiting coverage to a small fraction of available content. Most existing automatic AD generation methods frame the task as video clip captioning, requiring ground-truth timestamps and additional context cues such as character databases. Current benchmarks reinforce this framing, consisting of short video segments paired with automatic or task-mismatched annotations. We introduce StrAD, a benchmark for long-form AD generation on full-length videos spanning diverse genres such as movies, documentaries, short films, performances, and video games. We reformulate AD generation as streaming dense video captioning. Our approach processes full-length videos with a sliding window, inserting ADs into existing transcripts without ground-truth timestamps, and supports both fine-tuned models and zero-shot prompting of vision-language models. On the segment-level task with given timestamps, our fine-tuned StrAD-FT sets the state of the art on CMD-AD with 36.3 CIDEr (+10.0 over Shot-by-shot), establishes a reference point on StrAD (51.0 CIDEr), and remains competitive on MAD-Eval at 24.9 CIDEr. On the full-video streaming task, StrAD-FT reaches a SODA score of 2.4 against 1.1 for our zero-shot baseline StrAD-Zero, though both exhibit limitations in temporal localization and narrative coherence. While prior work has tackled full-video AD generation in an offline, multi-stage fashion, ours is the first streaming approach, generating ADs on the fly without ground-truth timestamps. StrAD makes progress on full-video AD generation measurable, a prerequisite for scaling accessibility.

cs.CV↗

arXiv2026

Dream-MPC: Gradient-Based Model Predictive Control with Latent Imagination

State-of-the-art model-based Reinforcement Learning (RL) approaches either use gradient-free, population-based methods for planning, learned policy networks, or a combination of policy networks and planning. Hybrid approaches that combine Model Predictive Control (MPC) with a learned model and a policy prior to leverage the advantages of both paradigms have shown promising results. However, these approaches typically rely on gradient-free optimization methods, which can be computationally expensive for high-dimensional control tasks. While gradient-based methods are a promising alternative, recent works have empirically shown that gradient-based methods often perform worse than their gradient-free counterparts. We propose Dream-MPC, a novel approach that generates few candidate trajectories from a rolled-out policy and optimizes each trajectory by gradient ascent using a learned world model, uncertainty regularization and amortization of optimization iterations over time by reusing previously optimized actions. Our results on 24 continuous control tasks show that Dream-MPC can significantly improve the performance of the underlying policy and can outperform gradient-free MPC and state-of-the-art baselines. Code and videos are available at https://dream-mpc.github.io.

cs.LG↗

arXiv2026

ContactFlow: A video action conditioning that transfers across embodiments

World models offer a promising route toward robot planning by enabling agents to imagine and verify the consequences of actions before execution. However, current video-based world models often struggle to capture the physical constraints that govern manipulation, particularly contact. Further, their action conditioning is often constrained to specific embodiments such as parallel grippers. We propose \emph{Contact Flow}, an embodiment-agnostic action representation that encodes manipulation through the trajectory of 3D contact points between an actor and a target object. By discarding actor-specific appearance and kinematics, Contact Flow provides a shared conditioning signal for both human demonstrations and robotic execution. Therefore, we can train a large-scale video generative model on both human and robotic object interaction videos conditioned on Contact Flow, yielding a world model that predicts physically plausible manipulation outcomes. We integrate this model into a propose-imagine-verify-act pipeline, where generated rollouts are assessed by a vision-language model before execution. Experiments on the DROID dataset and real-world tabletop manipulation tasks demonstrate that Contact Flow enables transfer between human demonstrations and different robotic embodiments.

cs.RO↗

arXiv2026

Slot-MPC: Goal-Conditioned Model Predictive Control with Object-Centric Representations

Predictive world models enable agents to model scene dynamics and reason about the consequences of their actions. Inspired by human perception, object-centric world models capture scene dynamics using object-level representations, which can be used for downstream applications such as action planning. However, most object-centric world models and reinforcement learning (RL) approaches learn reactive policies that are fixed at inference time, limiting generalization to novel situations. We propose Slot-MPC, an object-centric world modeling framework that enables planning through Model Predictive Control (MPC). Slot-MPC leverages vision encoders to learn slot-based representations, which encode individual objects in the scene, and uses these structured representations to learn an action-conditioned object-centric dynamics model. At inference time, the learned dynamics model enables action planning via MPC, allowing agents to adapt to previously unseen situations. Since the learned world model is differentiable, we can use gradient-based MPC to directly optimize actions, which is computationally more efficient than relying on gradient-free, sampling-based MPC methods. Experiments on simulated robotic manipulation tasks show that Slot-MPC improves both task performance and planning efficiency compared to non-object-centric world model baselines. In the considered offline setting with limited state-action coverage, we find that gradient-based MPC performs better than gradient-free, sampling-based MPC. Our results demonstrate that explicitly structured, object-centric representations provide a strong inductive bias for controllable and generalizable decision-making. Code and additional results are available at https://slot-mpc.github.io.

cs.LG↗

arXiv2026

EL3DD: Extended Latent 3D Diffusion for Language Conditioned Multitask Manipulation

Acting in human environments is a crucial capability for general-purpose robots, necessitating a robust understanding of natural language and its application to physical tasks. This paper seeks to harness the capabilities of diffusion models within a visuomotor policy framework that merges visual and textual inputs to generate precise robotic trajectories. By employing reference demonstrations during training, the model learns to execute manipulation tasks specified through textual commands within the robot's immediate environment. The proposed research aims to extend an existing model by leveraging improved embeddings, and adapting techniques from diffusion models for image generation. We evaluate our methods on the CALVIN dataset, proving enhanced performance on various manipulation tasks and an increased long-horizon success rate when multiple tasks are executed in sequence. Our approach reinforces the usefulness of diffusion models and contributes towards general multitask manipulation.

cs.RO↗

arXiv2026

Efficient Image Annotation via Semi-Supervised Object Segmentation with Label Propagation

Reliable object perception is necessary for general-purpose service robots. Open-vocabulary detectors struggle to generalize beyond a few classes and fully supervised training of object detectors requires time-intensive annotations. We present a semi-supervised label propagation approach for household object segmentation. A segment proposer generates class-agnostic masks, and an ensemble of Hopfield networks assigns labels by learning representative embeddings in complementary foundation model embedding spaces (CLIP, ViT, Theia). Our approach scales to 50 object classes with limited annotation overhead and can automatically label 60% of the data in a RoboCup@Home setting, where preparation time is severely constrained. Dataset and code are publicly available at https://github.com/ais-bonn/label_propagation.

cs.CV↗

arXiv2026

Perception-aware Exploration for Consumer-grade UAVs

In our work, we extend the current state-of-the-art approach for autonomous multi-UAV exploration to consumer-level UAVs, such as the DJI Mini 3 Pro. We propose a pipeline that selects viewpoint pairs from which the depth can be estimated and plans the trajectory that satisfies motion constraints necessary for odometry estimation. For the multi-UAV exploration, we propose a semi-distributed communication scheme that distributes the workload in a balanced manner. We evaluate our model performance in simulation for different numbers of UAVs and prove its ability to safely explore the environment and reconstruct the map even with the hardware limitations of consumer-grade UAVs.

cs.RO↗

arXiv2026

MonoEM-GS: Monocular Expectation-Maximization Gaussian Splatting SLAM

Feed-forward geometric foundation models can infer dense point clouds and camera motion directly from RGB streams, providing priors for monocular SLAM. However, their predictions are often view-dependent and noisy: geometry can vary across viewpoints and under image transformations, and local metric properties may drift between frames. We present MonoEM-GS, a monocular mapping pipeline that integrates such geometric predictions into a global Gaussian Splatting representation while explicitly addressing these inconsistencies. MonoEM-GS couples Gaussian Splatting with an Expectation--Maximization formulation to stabilize geometry, and employs ICP-based alignment for monocular pose estimation. Beyond geometry, MonoEM-GS parameterizes Gaussians with multi-modal features, enabling in-place open-set segmentation and other downstream queries directly on the reconstructed map. We evaluate MonoEM-GS on 7-Scenes, TUM RGB-D and Replica, and compare against recent baselines.

cs.RO↗

arXiv2026

OMCL: Open-vocabulary Monte Carlo Localization

Robust robot localization is an important prerequisite for navigation, but it becomes challenging when the map and robot measurements are obtained from different sensors. Prior methods are often tailored to specific environments, relying on closed-set semantics or fine-tuned features. In this work, we extend Monte Carlo Localization with vision-language features, allowing OMCL to robustly compute the likelihood of visual observations given a camera pose and a 3D map created from posed RGB-D images or aligned point clouds. These open-vocabulary features enable us to associate observations and map elements from different modalities, and to natively initialize global localization through natural language descriptions of nearby objects. We evaluate our approach using Matterport3D and Replica for indoor scenes and demonstrate generalization on SemanticKITTI for outdoor scenes.

cs.RO↗

arXiv2026

TextOCVP: Object-Centric Video Prediction with Language Guidance

Understanding and forecasting future scene states is critical for autonomous agents to plan and act effectively in complex environments. Object-centric models, with structured latent spaces, have shown promise in modeling object dynamics and predicting future scene states, but often struggle to scale beyond simple synthetic datasets and to integrate external guidance, limiting their applicability in robotics. To address these limitations, we propose TextOCVP, an object-centric model for video prediction guided by textual descriptions. TextOCVP parses an observed scene into object representations, called slots, and utilizes a text-conditioned transformer predictor to forecast future object states and video frames. Our approach jointly models object dynamics and interactions while incorporating textual guidance, enabling accurate and controllable predictions. TextOCVP's structured latent space offers a more precise control of the forecasting process, outperforming several video prediction baselines on two datasets. Additionally, we show that structured object-centric representations provide superior robustness to novel scene configurations, as well as improved controllability and interpretability, enabling more precise and understandable predictions. Videos and code are available at https://play-slot.github.io/TextOCVP.

cs.CV↗

arXiv2025

Aerial Assistance System for Automated Firefighting during Turntable Ladder Operations

Fires in industrial facilities pose special challenges to firefighters, e.g., due to the sheer size and scale of the buildings. The resulting visual obstructions impair firefighting accuracy, further compounded by inaccurate assessments of the fire's location. Such imprecision simultaneously increases the overall damage and prolongs the fire-brigades operation unnecessarily. We propose an automated assistance system for firefighting using a motorized fire monitor on a turntable ladder with aerial support from an unmanned aerial vehicle (UAV). The UAV flies autonomously within an obstacle-free flight funnel derived from geodata, detecting and localizing heat sources. An operator supervises the operation on a handheld controller and selects a fire target in reach. After the selection, the UAV automatically plans and traverses between two triangulation poses for continued fire localization. Simultaneously, our system steers the fire monitor to ensure the water jet reaches the detected heat source. In preliminary tests, our assistance system successfully localized multiple heat sources and directed a water jet towards the fires.

cs.RO↗

arXiv2025

LIO-MARS: Non-uniform Continuous-time Trajectories for Real-time LiDAR-Inertial-Odometry

Autonomous robotic systems heavily rely on environment knowledge to safely navigate. For search & rescue, a flying robot requires robust real-time perception, enabled by complementary sensors. IMU data constrains acceleration and rotation, whereas LiDAR measures accurate distances around the robot. Building upon the LiDAR odometry MARS, our LiDAR-inertial odometry (LIO) jointly aligns multi-resolution surfel maps with a Gaussian mixture model (GMM) using a continuous-time B-spline trajectory. Our new scan window uses non-uniform temporal knot placement to ensure continuity over the whole trajectory without additional scan delay. Moreover, we accelerate essential covariance and GMM computations with Kronecker sums and products by a factor of 3.3. An unscented transform de-skews surfels, while a splitting into intra-scan segments facilitates motion compensation during spline optimization. Complementary soft constraints on relative poses and preintegrated IMU pseudo-measurements further improve robustness and accuracy. Extensive evaluation showcases the state-of-the-art quality of our LIO-MARS w.r.t. recent LIO systems on various handheld, ground and aerial vehicle-based datasets.

cs.RO↗

arXiv2025

Scaling Laws for Conditional Emergence of Multilingual Image Captioning via Generalization from Translation

Cross-lingual, cross-task transfer is challenged by task-specific data scarcity, which becomes more severe as language support grows and is further amplified in vision-language models (VLMs). We investigate multilingual generalization in encoder-decoder transformer VLMs to enable zero-shot image captioning in languages encountered only in the translation task. In this setting, the encoder must learn to generate generalizable, task-aware latent vision representations to instruct the decoder via inserted cross-attention layers. To analyze scaling behavior, we train Florence-2 based and Gemma-2 based models (0.4B to 11.2B parameters) on a synthetic dataset using varying compute budgets. While all languages in the dataset have image-aligned translations, only a subset of them include image captions. Notably, we show that captioning can emerge using a language prefix, even when this language only appears in the translation task. We find that indirect learning of unseen task-language pairs adheres to scaling laws that are governed by the multilinguality of the model, model size, and seen training samples. Finally, we demonstrate that the scaling laws extend to downstream tasks, achieving competitive performance through fine-tuning in multimodal machine translation (Multi30K, CoMMuTE), lexical disambiguation (CoMMuTE), and image captioning (Multi30K, XM3600, COCO Karpathy).

cs.CL↗

arXiv2025

Remote Autonomy for Multiple Small Lowcost UAVs in GNSS-denied Search and Rescue Operations

In recent years, consumer-grade UAVs have been widely adopted by first responders. In general, they are operated manually, which requires trained pilots, especially in unknown GNSS-denied environments and in the vicinity of structures. Autonomous flight can facilitate the application of UAVs and reduce operator strain. However, autonomous systems usually require special programming interfaces, custom sensor setups, and strong onboard computers, which limits a broader deployment. We present a system for autonomous flight using lightweight consumer-grade DJI drones. They are controlled by an Android app for state estimation and obstacle avoidance directly running on the UAV's remote control. Our ground control station enables a single operator to configure and supervise multiple heterogeneous UAVs at once. Furthermore, it combines the observations of all UAVs into a joint 3D environment model for improved situational awareness.

cs.RO↗

arXiv2025

Foveated Compression for Immersive Telepresence Visualization

Immersive televisualization is important both for telepresence and teleoperation, but resolution and fidelity are often limited by communication bandwidth constraints. We propose a lightweight method for foveated compression of immersive televisualization video streams that can be easily integrated with common video codecs, reducing the required bandwidth if eye tracking data is available. Specifically, we show how to spatially adjust the Quantization Parameter of modern block-based video codecs in a adaptive way based on eye tracking information. The foveal region is transmitted with high fidelity while quality is reduced in the peripheral region, saving bandwidth. We integrate our method with the NimbRo avatar system, which won the ANA Avatar XPRIZE competition. Our experiments show that bandwidth can be reduced to a third without sacrificing immersion. We analyze transmission fidelity with qualitative examples and report quantitative results.

eess.IV↗

arXiv2025

Leveraging Vision-Language Models for Open-Vocabulary Instance Segmentation and Tracking

Vision-language models (VLMs) excel in visual understanding but often lack reliable grounding capabilities and actionable inference rates. Integrating them with open-vocabulary object detection (OVD), instance segmentation, and tracking leverages their strengths while mitigating these drawbacks. We utilize VLM-generated structured descriptions to identify visible object instances, collect application-relevant attributes, and inform an open-vocabulary detector to extract corresponding bounding boxes that are passed to a video segmentation model providing segmentation masks and tracking. Once initialized, this model directly extracts segmentation masks, processing image streams in real time with minimal computational overhead. Tracks can be updated online as needed by generating new structured descriptions and detections. This combines the descriptive power of VLMs with the grounding capability of OVD and the pixel-level understanding and speed of video segmentation. Our evaluation across datasets and robotics platforms demonstrates the broad applicability of this approach, showcasing its ability to extract task-specific attributes from non-standard objects in dynamic environments. Code, data, videos, and benchmarks are available at https://vlm-gist.github.io

cs.CV↗

arXiv2025

Iterative Motion Compensation for Canonical 3D Reconstruction from UAV Plant Images Captured in Windy Conditions

3D phenotyping of plants plays a crucial role for understanding plant growth, yield prediction, and disease control. We present a pipeline capable of generating high-quality 3D reconstructions of individual agricultural plants. To acquire data, a small commercially available UAV captures images of a selected plant. Apart from placing ArUco markers, the entire image acquisition process is fully autonomous, controlled by a self-developed Android application running on the drone's controller. The reconstruction task is particularly challenging due to environmental wind and downwash of the UAV. Our proposed pipeline supports the integration of arbitrary state-of-the-art 3D reconstruction methods. To mitigate errors caused by leaf motion during image capture, we use an iterative method that gradually adjusts the input images through deformation. Motion is estimated using optical flow between the original input images and intermediate 3D reconstructions rendered from the corresponding viewpoints. This alignment gradually reduces scene motion, resulting in a canonical representation. After a few iterations, our pipeline improves the reconstruction of state-of-the-art methods and enables the extraction of high-resolution 3D meshes. We will publicly release the source code of our reconstruction pipeline. Additionally, we provide a dataset consisting of multiple plants from various crops, captured across different points in time.

cs.CV↗