Stefan Leutenegger

Christopher L. Choi, Binbin Xu, Stefan Leutenegger

Visual-Inertial (VI) sensors are popular in robotics, self-driving vehicles, and augmented and virtual reality applications. In order to use them for any computer vision or state-estimation task, a good calibration is essential. However, collecting informative calibration data in order to render the calibration parameters observable is not trivial for a non-expert. In this work, we introduce a novel VI calibration pipeline that guides a non-expert with the use of a graphical user interface and information theory in collecting informative calibration data with Next-Best-View and Next-Best-Trajectory suggestions to calibrate the intrinsics, extrinsics, and temporal misalignment of a VI sensor. We show through experiments that our method is faster, more accurate, and more consistent than state-of-the-art alternatives. Specifically, we show how calibrations with our proposed method achieve higher accuracy estimation results when used by state-of-the-art VI Odometry as well as VI-SLAM approaches. The source code of our software can be found on: https://github.com/chutsu/yac.

Marco Camurri, Enrico Tomelleri, Matías Mattamala et al. · oxford

Covering one third of Earth's land surface, forests are vital to global biodiversity, climate regulation, and human well-being. In Europe, forests and woodlands reach approximately 40% of land area, and the forestry sector is central to achieving the EU's climate neutrality and biodiversity goals; these emphasize sustainable forest management, increased use of long-lived wood products, and resilient forest ecosystems. To meet these goals and properly address their inherent challenges, current practices require further innovation. This chapter introduces DigiForest, a novel, large-scale precision forestry approach leveraging digital technologies and autonomous robotics. DigiForest is structured around four main components: (1) autonomous, heterogeneous mobile robots (aerial, legged, and marsupial) for tree-level data collection; (2) automated extraction of tree traits to build forest inventories; (3) a Decision Support System (DSS) for forecasting forest growth and supporting decision-making; and (4) low-impact selective logging using purpose-built autonomous harvesters. These technologies have been extensively validated in real-world conditions in several locations, including forests in Finland, the UK, and Switzerland.

Jiaxin Wei, Stefan Leutenegger, Laurent Kneip

Perspective-$n$-Point (P$n$P) stands as a fundamental algorithm for pose estimation in various applications. In this paper, we present a new approach to the P$n$P problem with relaxed constraints, eliminating the need for precise 3D coordinates, which is especially suitable for object pose estimation where corresponding object models may not be available in practice. Built upon the classical EP$n$P solver, we refer to it as AEP$n$P due to its ability to handle unknown anisotropic scaling factors in addition to the common 6D transformation. Through a few algebraic manipulations and a well-chosen frame of reference, this new problem can be boiled down to a simple linear null-space problem followed by point registration-based identification of a similarity transformation. Experimental results on both simulated and real datasets demonstrate the effectiveness of AEP$n$P as a flexible and practical solution to object pose estimation. Code: https://github.com/goldoak/AEPnP.

Binbin Xu, Andrew J. Davison, Stefan Leutenegger

In this paper, we propose a novel object-level mapping system that can simultaneously segment, track, and reconstruct objects in dynamic scenes. It can further predict and complete their full geometries by conditioning on reconstructions from depth inputs and a category-level shape prior with the aim that completed object geometry leads to better object reconstruction and tracking accuracy. For each incoming RGB-D frame, we perform instance segmentation to detect objects and build data associations between the detection and the existing object maps. A new object map will be created for each unmatched detection. For each matched object, we jointly optimise its pose and latent geometry representations using geometric residual and differential rendering residual towards its shape prior and completed geometry. Our approach shows better tracking and reconstruction performance compared to methods using traditional volumetric mapping or learned shape prior approaches. We evaluate its effectiveness by quantitatively and qualitatively testing it in both synthetic and real-world sequences.

Tristan Laidlow, Jan Czarnowski, Stefan Leutenegger

While the keypoint-based maps created by sparse monocular simultaneous localisation and mapping (SLAM) systems are useful for camera tracking, dense 3D reconstructions may be desired for many robotic tasks. Solutions involving depth cameras are limited in range and to indoor spaces, and dense reconstruction systems based on minimising the photometric error between frames are typically poorly constrained and suffer from scale ambiguity. To address these issues, we propose a 3D reconstruction system that leverages the output of a convolutional neural network (CNN) to produce fully dense depth maps for keyframes that include metric scale. Our system, DeepFusion, is capable of producing real-time dense reconstructions on a GPU. It fuses the output of a semi-dense multiview stereo algorithm with the depth and gradient predictions of a CNN in a probabilistic fashion, using learned uncertainties produced by the network. While the network only needs to be run once per keyframe, we are able to optimise for the depth map with each new frame so as to constantly make use of new geometric constraints. Based on its performance on synthetic and real-world datasets, we demonstrate that DeepFusion is capable of performing at least as well as other comparable systems.

Tristan Laidlow, Michael Bloesch, Wenbin Li et al.

While dense visual SLAM methods are capable of estimating dense reconstructions of the environment, they suffer from a lack of robustness in their tracking step, especially when the optimisation is poorly initialised. Sparse visual SLAM systems have attained high levels of accuracy and robustness through the inclusion of inertial measurements in a tightly-coupled fusion. Inspired by this performance, we propose the first tightly-coupled dense RGB-D-inertial SLAM system. Our system has real-time capability while running on a GPU. It jointly optimises for the camera pose, velocity, IMU biases and gravity direction while building up a globally consistent, fully dense surfel-based 3D reconstruction of the environment. Through a series of experiments on both synthetic and real world datasets, we show that our dense visual-inertial SLAM system is more robust to fast motions and periods of low texture and low geometric variation than a related RGB-D-only SLAM system.

Dorian F. Henning, Tristan Laidlow, Stefan Leutenegger

Estimating human motion from video is an active research area due to its many potential applications. Most state-of-the-art methods predict human shape and posture estimates for individual images and do not leverage the temporal information available in video. Many "in the wild" sequences of human motion are captured by a moving camera, which adds the complication of conflated camera and human motion to the estimation. We therefore present BodySLAM, a monocular SLAM system that jointly estimates the position, shape, and posture of human bodies, as well as the camera trajectory. We also introduce a novel human motion model to constrain sequential body postures and observe the scale of the scene. Through a series of experiments on video sequences of human motion captured by a moving monocular camera, we demonstrate that BodySLAM improves estimates of all human body parameters and camera poses when compared to estimating these separately.

Yuxuan Xue, Haolong Li, Stefan Leutenegger et al.

Visual reconstruction of fast non-rigid object deformations over time is a challenge for conventional frame-based cameras. In this paper, we propose a novel approach for reconstructing such deformations using measurements from event-based cameras. Under the assumption of a static background, where all events are generated by the motion, our approach estimates the deformation of objects from events generated at the object contour in a probabilistic optimization framework. It associates events to mesh faces on the contour and maximizes the alignment of the line of sight through the event pixel with the associated face. In experiments on synthetic and real data, we demonstrate the advantages of our method over state-of-the-art optimization and learning-based approaches for reconstructing the motion of human hands. A video of the experiments is available at https://youtu.be/gzfw7i5OKjg

Yifei Ren, Binbin Xu, Christopher L. Choi et al.

In this paper, we present a tightly-coupled visual-inertial object-level multi-instance dynamic SLAM system. Even in extremely dynamic scenes, it can robustly optimise for the camera pose, velocity, IMU biases and build a dense 3D reconstruction object-level map of the environment. Our system can robustly track and reconstruct the geometries of arbitrary objects, their semantics and motion by incrementally fusing associated colour, depth, semantic, and foreground object probabilities into each object model thanks to its robust sensor and object tracking. In addition, when an object is lost or moved outside the camera field of view, our system can reliably recover its pose upon re-observation. We demonstrate the robustness and accuracy of our method by quantitatively and qualitatively testing it in real-world data sequences.

Anna Penzkofer, Simon Schaefer, Florian Strohm et al.

While deep reinforcement learning (RL) agents outperform humans on an increasing number of tasks, training them requires data equivalent to decades of human gameplay. Recent hierarchical RL methods have increased sample efficiency by incorporating information inherent to the structure of the decision problem but at the cost of having to discover or use human-annotated sub-goals that guide the learning process. We show that intentions of human players, i.e. the precursor of goal-oriented decisions, can be robustly predicted from eye gaze even for the long-horizon sparse rewards task of Montezuma's Revenge - one of the most challenging RL tasks in the Atari2600 game suite. We propose Int-HRL: Hierarchical RL with intention-based sub-goals that are inferred from human eye gaze. Our novel sub-goal extraction pipeline is fully automatic and replaces the need for manual sub-goal annotation by human experts. Our evaluations show that replacing hand-crafted sub-goals with automatically extracted intentions leads to a HRL agent that is significantly more sample efficient than previous methods.

Yingye Xin, Xingxing Zuo, Dongyue Lu et al.

We present a real-time visual-inertial dense mapping method capable of performing incremental 3D mesh reconstruction with high quality using only sequential monocular images and inertial measurement unit (IMU) readings. 6-DoF camera poses are estimated by a robust feature-based visual-inertial odometry (VIO), which also generates noisy sparse 3D map points as a by-product. We propose a sparse point aided multi-view stereo neural network (SPA-MVSNet) that can effectively leverage the informative but noisy sparse points from the VIO system. The sparse depth from VIO is firstly completed by a single-view depth completion network. This dense depth map, although naturally limited in accuracy, is then used as a prior to guide our MVS network in the cost volume generation and regularization for accurate dense depth prediction. Predicted depth maps of keyframe images by the MVS network are incrementally fused into a global map using TSDF-Fusion. We extensively evaluate both the proposed SPA-MVSNet and the entire visual-inertial dense mapping system on several public datasets as well as our own dataset, demonstrating the system's impressive generalization capabilities and its ability to deliver high-quality 3D mesh reconstruction online. Our proposed dense mapping system achieves a 39.7% improvement in F-score over existing systems when evaluated on the challenging scenarios of the EuRoC dataset.

Jiaxin Wei, Stefan Leutenegger

Traditional volumetric fusion algorithms preserve the spatial structure of 3D scenes, which is beneficial for many tasks in computer vision and robotics. However, they often lack realism in terms of visualization. Emerging 3D Gaussian splatting bridges this gap, but existing Gaussian-based reconstruction methods often suffer from artifacts and inconsistencies with the underlying 3D structure, and struggle with real-time optimization, unable to provide users with immediate feedback in high quality. One of the bottlenecks arises from the massive amount of Gaussian parameters that need to be updated during optimization. Instead of using 3D Gaussian as a standalone map representation, we incorporate it into a volumetric mapping system to take advantage of geometric information and propose to use a quadtree data structure on images to drastically reduce the number of splats initialized. In this way, we simultaneously generate a compact 3D Gaussian map with fewer artifacts and a volumetric map on the fly. Our method, GSFusion, significantly enhances computational efficiency without sacrificing rendering quality, as demonstrated on both synthetic and real datasets. Code will be available at https://github.com/goldoak/GSFusion.

Simon Schaefer, Dorian F. Henning, Stefan Leutenegger

An accurate and uncertainty-aware 3D human body pose estimation is key to enabling truly safe but efficient human-robot interactions. Current uncertainty-aware methods in 3D human pose estimation are limited to predicting the uncertainty of the body posture, while effectively neglecting the body shape and root pose. In this work, we present GloPro, which to the best of our knowledge the first framework to predict an uncertainty distribution of a 3D body mesh including its shape, pose, and root pose, by efficiently fusing visual clues with a learned motion model. We demonstrate that it vastly outperforms state-of-the-art methods in terms of human trajectory accuracy in a world coordinate system (even in the presence of severe occlusions), yields consistent uncertainty distributions, and can run in real-time.

Tristan Laidlow, Jan Czarnowski, Andrea Nicastro et al.

The best way to combine the results of deep learning with standard 3D reconstruction pipelines remains an open problem. While systems that pass the output of traditional multi-view stereo approaches to a network for regularisation or refinement currently seem to get the best results, it may be preferable to treat deep neural networks as separate components whose results can be probabilistically fused into geometry-based systems. Unfortunately, the error models required to do this type of fusion are not well understood, with many different approaches being put forward. Recently, a few systems have achieved good results by having their networks predict probability distributions rather than single values. We propose using this approach to fuse a learned single-view depth prior into a standard 3D reconstruction system. Our system is capable of incrementally producing dense depth maps for a set of keyframes. We train a deep neural network to predict discrete, nonparametric probability distributions for the depth of each pixel from a single image. We then fuse this "probability volume" with another probability volume based on the photometric consistency between subsequent frames and the keyframe image. We argue that combining the probability volumes from these two sources will result in a volume that is better conditioned. To extract depth maps from the volume, we minimise a cost function that includes a regularisation term based on network predicted surface normals and occlusion boundaries. Through a series of experiments, we demonstrate that each of these components improves the overall performance of the system.

Diego Hidalgo-Carvajal, Hanzhi Chen, Gemma C. Bettelani et al.

The progressive prevalence of robots in human-suited environments has given rise to a myriad of object manipulation techniques, in which dexterity plays a paramount role. It is well-established that humans exhibit extraordinary dexterity when handling objects. Such dexterity seems to derive from a robust understanding of object properties (such as weight, size, and shape), as well as a remarkable capacity to interact with them. Hand postures commonly demonstrate the influence of specific regions on objects that need to be grasped, especially when objects are partially visible. In this work, we leverage human-like object understanding by reconstructing and completing their full geometry from partial observations, and manipulating them using a 7-DoF anthropomorphic robot hand. Our approach has significantly improved the grasping success rates of baselines with only partial reconstruction by nearly 30% and achieved over 150 successful grasps with three different object categories. This demonstrates our approach's consistent ability to predict and execute grasping postures based on the completed object shapes from various directions and positions in real-world scenarios. Our work opens up new possibilities for enhancing robotic applications that require precise grasping and manipulation skills of real-world reconstructed objects.

Simon Schaefer, Joshua Näf, Stefan Leutenegger

Robust and accurate perception of humans in their 3D scene context is essential for integrating robots into everyday environments. Existing approaches, however, often fail to predict plausible and accurate human motion estimates that are consistent with the surrounding scene, especially in the presence of heavy occlusions or partial visibility. This can limit both safety and efficiency for robotic operations. We introduce HumanFlow, a latent diffusion model that unifies human motion tracking and forecasting, conditioned on the 3D scene context. We show that our human motion model produces smooth and accurate predictions under challenging conditions, including heavy occlusions, and outperforms state-of-the-art methods in tracking accuracy while being significantly more efficient. Furthermore, we show how HumanFlow's latent space can be tightly coupled with control by conditioning a flow-matching-based, approximate MPC policy on these representations. We validate our policy in simulation with real human trajectories for MAV social navigation, demonstrating superior navigation performance and remaining collision-free, even under partial observability of the human.

Daoyi Gao, Dávid Rozenberszki, Stefan Leutenegger et al.

Perceiving 3D structures from RGB images based on CAD model primitives can enable an effective, efficient 3D object-based representation of scenes. However, current approaches rely on supervision from expensive annotations of CAD models associated with real images, and encounter challenges due to the inherent ambiguities in the task -- both in depth-scale ambiguity in monocular perception, as well as inexact matches of CAD database models to real observations. We thus propose DiffCAD, the first weakly-supervised probabilistic approach to CAD retrieval and alignment from an RGB image. We formulate this as a conditional generative task, leveraging diffusion to learn implicit probabilistic models capturing the shape, pose, and scale of CAD objects in an image. This enables multi-hypothesis generation of different plausible CAD reconstructions, requiring only a few hypotheses to characterize ambiguities in depth/scale and inexact shape matches. Our approach is trained only on synthetic data, leveraging monocular depth and mask estimates to enable robust zero-shot adaptation to various real target domains. Despite being trained solely on synthetic data, our multi-hypothesis approach can even surpass the supervised state-of-the-art on the Scan2CAD dataset by 5.9% with 8 hypotheses.

Philipp Schoch, Fan Yang, Yuntao Ma et al.

Current visual navigation systems often treat the environment as static, lacking the ability to adaptively interact with obstacles. This limitation leads to navigation failure when encountering unavoidable obstructions. In response, we introduce IN-Sight, a novel approach to self-supervised path planning, enabling more effective navigation strategies through interaction with obstacles. Utilizing RGB-D observations, IN-Sight calculates traversability scores and incorporates them into a semantic map, facilitating long-range path planning in complex, maze-like environments. To precisely navigate around obstacles, IN-Sight employs a local planner, trained imperatively on a differentiable costmap using representation learning techniques. The entire framework undergoes end-to-end training within the state-of-the-art photorealistic Intel SPEAR Simulator. We validate the effectiveness of IN-Sight through extensive benchmarking in a variety of simulated scenarios and ablation studies. Moreover, we demonstrate the system's real-world applicability with zero-shot sim-to-real transfer, deploying our planner on the legged robot platform ANYmal, showcasing its practical potential for interactive navigation in real environments.

Boyang Sun, Hanzhi Chen, Stefan Leutenegger et al.

Exploration of unknown environments is crucial for autonomous robots; it allows them to actively reason and decide on what new data to acquire for different tasks, such as mapping, object discovery, and environmental assessment. Existing solutions, such as frontier-based exploration approaches, rely heavily on 3D map operations, which are limited by map quality and, more critically, often overlook valuable context from visual cues. This work aims at leveraging 2D visual cues for efficient autonomous exploration, addressing the limitations of extracting goal poses from a 3D map. We propose a visual-only frontier-based exploration system, with FrontierNet as its core component. FrontierNet is a learning-based model that (i) proposes frontiers, and (ii) predicts their information gain, from posed RGB images enhanced by monocular depth priors. Our approach provides an alternative to existing 3D-dependent goal-extraction approaches, achieving a 15\% improvement in early-stage exploration efficiency, as validated through extensive simulations and real-world experiments. The project is available at https://github.com/cvg/FrontierNet.

Joshua Knights, Sebastián Barbas Laina, Peyman Moghadam et al.

This paper proposes SOLVR, a unified pipeline for learning based LiDAR-Visual re-localisation which performs place recognition and 6-DoF registration across sensor modalities. We propose a strategy to align the input sensor modalities by leveraging stereo image streams to produce metric depth predictions with pose information, followed by fusing multiple scene views from a local window using a probabilistic occupancy framework to expand the limited field-of-view of the camera. Additionally, SOLVR adopts a flexible definition of what constitutes positive examples for different training losses, allowing us to simultaneously optimise place recognition and registration performance. Furthermore, we replace RANSAC with a registration function that weights a simple least-squares fitting with the estimated inlier likelihood of sparse keypoint correspondences, improving performance in scenarios with a low inlier ratio between the query and retrieved place. Our experiments on the KITTI and KITTI360 datasets show that SOLVR achieves state-of-the-art performance for LiDAR-Visual place recognition and registration, particularly improving registration accuracy over larger distances between the query and retrieved place.

Jiaxin Wei, Stefan Leutenegger, Simon Schaefer

Recent developments in 3D Gaussian Splatting have significantly enhanced novel view synthesis, yet generating high-quality renderings from extreme novel viewpoints or partially observed regions remains challenging. Meanwhile, diffusion models exhibit strong generative capabilities, but their reliance on text prompts and lack of awareness of specific scene information hinder accurate 3D reconstruction tasks. To address these limitations, we introduce GSFix3D, a novel framework that improves the visual fidelity in under-constrained regions by distilling prior knowledge from diffusion models into 3D representations, while preserving consistency with observed scene details. At its core is GSFixer, a latent diffusion model obtained via our customized fine-tuning protocol that can leverage both mesh and 3D Gaussians to adapt pretrained generative models to a variety of environments and artifact types from different reconstruction methods, enabling robust novel view repair for unseen camera poses. Moreover, we propose a random mask augmentation strategy that empowers GSFixer to plausibly inpaint missing regions. Experiments on challenging benchmarks demonstrate that our GSFix3D and GSFixer achieve state-of-the-art performance, requiring only minimal scene-specific fine-tuning on captured data. Real-world test further confirms its resilience to potential pose errors. Our code and data will be made publicly available. Project page: https://gsfix3d.github.io.

Yannick Burkhardt, Simon Schaefer, Stefan Leutenegger

Event-based keypoint detection and matching holds significant potential, enabling the integration of event sensors into highly optimized Visual SLAM systems developed for frame cameras over decades of research. Unfortunately, existing approaches struggle with the motion-dependent appearance of keypoints and the complex noise prevalent in event streams, resulting in severely limited feature matching capabilities and poor performance on downstream tasks. To mitigate this problem, we propose SuperEvent, a data-driven approach to predict stable keypoints with expressive descriptors. Due to the absence of event datasets with ground truth keypoint labels, we leverage existing frame-based keypoint detectors on readily available event-aligned and synchronized gray-scale frames for self-supervision: we generate temporally sparse keypoint pseudo-labels considering that events are a product of both scene appearance and camera motion. Combined with our novel, information-rich event representation, we enable SuperEvent to effectively learn robust keypoint detection and description in event streams. Finally, we demonstrate the usefulness of SuperEvent by its integration into a modern sparse keypoint and descriptor-based SLAM framework originally developed for traditional cameras, surpassing the state-of-the-art in event-based SLAM by a wide margin. Source code is available at https://ethz-mrl.github.io/SuperEvent/.

Gang Chen, Sebastián Barbas Laina, Stefan Leutenegger et al.

Scene graph alignment establishes object correspondences between two 3D scene graphs constructed from partially overlapping observations. This enables efficient scene understanding and object-level relocalization when a robot revisits a place, as well as global map fusion across multiple agents. Such capabilities are essential for robots that require long-term memory for long-horizon tasks involving interactions with the environment. Existing approaches mainly focus on subscan-to-subscan (S2S) alignment and depend heavily on geometric point-cloud features, leaving frame-to-scan (F2S) alignment and open-set vision-language features underexplored. In addition, existing datasets for scene graph alignment remain small-scale with limited object diversity, constraining systematic training and evaluation. We present a unified and efficient scene graph alignment framework that predicts object correspondences by fusing vision-language, textual, and geometric features with spatial context. The framework comprises modules such as a distance-gated spatial attention encoder, a minimum-cost-flow-based allocator, and a global scene embedding generator to achieve accurate alignment even under large coordinate discrepancies. We further introduce ScanNet-SG, a large-scale dataset generated via an automated annotation pipeline with over 700k samples, covering 509 object categories from ScanNet labels and over 3k categories from GPT-4o-based tagging. Experiments show that our method achieves the best overall performance on both F2S and S2S tasks, substantially outperforming existing scene graph alignment methods. Our code and dataset are released at: https://autonomousrobots.nl/paper_websites/opensga.

Yannick Burkhardt, Sebastián Barbas Laina, Simon Boche et al.

The robustness of event cameras to high dynamic range and motion blur holds the potential to improve visual odometry systems in challenging environments. Although their high temporal resolution does not require synchronous processing, most event-based odometry methods still run at fixed rates, which simplifies system design but restricts latency and throughput. In this work, we present AERO-VIS, a stereo event-inertial SLAM system with an integrated, data-driven, robust, and performance-optimized keypoint detector. By processing the event stream asynchronously, the system dynamically adapts to downstream runtime demands, ensuring low-latency and real-time performance. When deploying AERO-VIS on a UAV, we achieve unprecedented accuracy in onboard event-based SLAM. These unique characteristics enable us to present the first purely event-based inertial SLAM system that demonstrates closed-loop UAV control and large-scale state estimation while relying solely on onboard compute. A video of the experiments and the source code are available at ethz-mrl.github.io/AERO-VIS.

Jens Naumann, Binbin Xu, Stefan Leutenegger et al.

We introduce a novel monocular visual odometry (VO) system, NeRF-VO, that integrates learning-based sparse visual odometry for low-latency camera tracking and a neural radiance scene representation for fine-detailed dense reconstruction and novel view synthesis. Our system initializes camera poses using sparse visual odometry and obtains view-dependent dense geometry priors from a monocular prediction network. We harmonize the scale of poses and dense geometry, treating them as supervisory cues to train a neural implicit scene representation. NeRF-VO demonstrates exceptional performance in both photometric and geometric fidelity of the scene representation by jointly optimizing a sliding window of keyframed poses and the underlying dense geometry, which is accomplished through training the radiance field with volume rendering. We surpass SOTA methods in pose estimation accuracy, novel view synthesis fidelity, and dense reconstruction quality across a variety of synthetic and real-world datasets while achieving a higher camera tracking frequency and consuming less GPU memory.

Hanfeng Wu, Xingxing Zuo, Stefan Leutenegger et al.

We introduce DyNFL, a novel neural field-based approach for high-fidelity re-simulation of LiDAR scans in dynamic driving scenes. DyNFL processes LiDAR measurements from dynamic environments, accompanied by bounding boxes of moving objects, to construct an editable neural field. This field, comprising separately reconstructed static background and dynamic objects, allows users to modify viewpoints, adjust object positions, and seamlessly add or remove objects in the re-simulated scene. A key innovation of our method is the neural field composition technique, which effectively integrates reconstructed neural assets from various scenes through a ray drop test, accounting for occlusions and transparent surfaces. Our evaluation with both synthetic and real-world environments demonstrates that DyNFL substantially improves dynamic scene LiDAR simulation, offering a combination of physical fidelity and flexible editing capabilities.

Hanzhi Chen, Boyang Sun, Anran Zhang et al.

Future robots are envisioned as versatile systems capable of performing a variety of household tasks. The big question remains, how can we bridge the embodiment gap while minimizing physical robot learning, which fundamentally does not scale well. We argue that learning from in-the-wild human videos offers a promising solution for robotic manipulation tasks, as vast amounts of relevant data already exist on the internet. In this work, we present VidBot, a framework enabling zero-shot robotic manipulation using learned 3D affordance from in-the-wild monocular RGB-only human videos. VidBot leverages a pipeline to extract explicit representations from them, namely 3D hand trajectories from videos, combining a depth foundation model with structure-from-motion techniques to reconstruct temporally consistent, metric-scale 3D affordance representations agnostic to embodiments. We introduce a coarse-to-fine affordance learning model that first identifies coarse actions from the pixel space and then generates fine-grained interaction trajectories with a diffusion model, conditioned on coarse actions and guided by test-time constraints for context-aware interaction planning, enabling substantial generalization to novel scenes and embodiments. Extensive experiments demonstrate the efficacy of VidBot, which significantly outperforms counterparts across 13 manipulation tasks in zero-shot settings and can be seamlessly deployed across robot systems in real-world environments. VidBot paves the way for leveraging everyday human videos to make robot learning more scalable.

Hanzhi Chen, Binbin Xu, Stefan Leutenegger

We present FuncGrasp, a framework that can infer dense yet reliable grasp configurations for unseen objects using one annotated object and single-view RGB-D observation via categorical priors. Unlike previous works that only transfer a set of grasp poses, FuncGrasp aims to transfer infinite configurations parameterized by an object-centric continuous grasp function across varying instances. To ease the transfer process, we propose Neural Surface Grasping Fields (NSGF), an effective neural representation defined on the surface to densely encode grasp configurations. Further, we exploit function-to-function transfer using sphere primitives to establish semantically meaningful categorical correspondences, which are learned in an unsupervised fashion without any expert knowledge. We showcase the effectiveness through extensive experiments in both simulators and the real world. Remarkably, our framework significantly outperforms several strong baseline methods in terms of density and reliability for generated grasps.

Sebastián Barbas Laina, Simon Boche, Sotiris Papatheodorou et al.

Geometrically accurate and semantically expressive map representations have proven invaluable to facilitate robust and safe mobile robot navigation and task planning. Nevertheless, real-time, open-vocabulary semantic understanding of large-scale unknown environments is still an open problem. In this paper we present FindAnything, an open-world mapping and exploration framework that incorporates vision-language information into dense volumetric submaps. Thanks to the use of vision-language features, FindAnything bridges the gap between pure geometric and open-vocabulary semantic information for a higher level of understanding while allowing to explore any environment without the help of any external source of ground-truth pose information. We represent the environment as a series of volumetric occupancy submaps, resulting in a robust and accurate map representation that deforms upon pose updates when the underlying SLAM system corrects its drift, allowing for a locally consistent representation between submaps. Pixel-wise vision-language features are aggregated from efficient SAM (eSAM)-generated segments, which are in turn integrated into object-centric volumetric submaps, providing a mapping from open-vocabulary queries to 3D geometry that is scalable also in terms of memory usage. The open-vocabulary map representation of FindAnything achieves state-of-the-art semantic accuracy in closed-set evaluations on the Replica dataset. This level of scene understanding allows a robot to explore environments based on objects or areas of interest selected via natural language queries. Our system is the first of its kind to be deployed on resource-constrained devices, such as MAVs, leveraging vision-language information for real-world robotic tasks.

Débora N. P. Oliveira, Joshua Knights, Sebastián Barbas Laina et al.

Loop closures are essential for correcting odometry drift and creating consistent maps, especially in the context of large-scale navigation. Current methods using dense point clouds for accurate place recognition do not scale well due to computationally expensive scan-to-scan comparisons. Alternative object-centric approaches are more efficient but often struggle with sensitivity to viewpoint variation. In this work, we introduce REGRACE, a novel approach that addresses these challenges of scalability and perspective difference in re-localization by using LiDAR-based submaps. We introduce rotation-invariant features for each labeled object and enhance them with neighborhood context through a graph neural network. To identify potential revisits, we employ a scalable bag-of-words approach, pooling one learned global feature per submap. Additionally, we define a revisit with geometrical consistency cues rather than embedding distance, allowing us to recognize far-away loop closures. Our evaluations demonstrate that REGRACE achieves similar results compared to state-of-the-art place recognition and registration baselines while being twice as fast. Code and models are publicly available.

Yao Zhong, Hanzhi Chen, Simon Schaefer et al.

Robots are expected to serve as intelligent assistants, helping humans with everyday household organization. A central challenge in this setting is the task of object placement, which requires reasoning about both semantic preferences (e.g., common-sense object relations) and geometric feasibility (e.g., collision avoidance). We present GOPLA, a hierarchical framework that learns generalizable object placement from augmented human demonstrations. A multi-modal large language model translates human instructions and visual inputs into structured plans that specify pairwise object relationships. These plans are then converted into 3D affordance maps with geometric common sense by a spatial mapper, while a diffusion-based planner generates placement poses guided by test-time costs, considering multi-plan distributions and collision avoidance. To overcome data scarcity, we introduce a scalable pipeline that expands human placement demonstrations into diverse synthetic training data. Extensive experiments show that our approach improves placement success rates by 30.04 percentage points over the runner-up, evaluated on positioning accuracy and physical plausibility, demonstrating strong generalization across a wide range of real-world robotic placement scenarios.

Hannah Schieber, Dominik Frischmann, Victor Schaack et al.

Mobile reconstruction has the potential to support time-critical tasks such as tele-guidance and disaster response, where operators must quickly gain an accurate understanding of the environment. Full high-fidelity scene reconstruction is computationally expensive and often unnecessary when only specific points of interest (POIs) matter for timely decision making. We address this challenge with CoRe-GS, a semantic POI-focused extension of Gaussian Splatting (GS). Instead of optimizing every scene element uniformly, CoRe-GS first produces a fast segmentation-ready GS representation and then selectively refines splats belonging to semantically relevant POIs detected during data acquisition. This targeted refinement reduces training time to 25\% compared to full semantic GS while improving novel view synthesis quality in the areas that matter most. We validate CoRe-GS on both real-world (SCRREAM) and synthetic (NeRDS 360) datasets, demonstrating that prioritizing POIs enables faster and higher-quality mobile reconstruction tailored to operational needs.

Theresa Huber, Simon Schaefer, Stefan Leutenegger

The assumption of a static environment is common in many geometric computer vision tasks like SLAM but limits their applicability in highly dynamic scenes. Since these tasks rely on identifying point correspondences between input images within the static part of the environment, we propose a graph neural network-based sparse feature matching network designed to perform robust matching under challenging conditions while excluding keypoints on moving objects. We employ a similar scheme of attentional aggregation over graph edges to enhance keypoint representations as state-of-the-art feature-matching networks but augment the graph with epipolar and temporal information and vastly reduce the number of graph edges. Furthermore, we introduce a self-supervised training scheme to extract pseudo labels for image pairs in dynamic environments from exclusively unprocessed visual-inertial data. A series of experiments show the superior performance of our network as it excludes keypoints on moving objects compared to state-of-the-art feature matching networks while still achieving similar results regarding conventional matching metrics. When integrated into a SLAM system, our network significantly improves performance, especially in highly dynamic scenes.

Dorian F. Henning, Christopher Choi, Simon Schaefer et al.

Robust, fast, and accurate human state - 6D pose and posture - estimation remains a challenging problem. For real-world applications, the ability to estimate the human state in real-time is highly desirable. In this paper, we present BodySLAM++, a fast, efficient, and accurate human and camera state estimation framework relying on visual-inertial data. BodySLAM++ extends an existing visual-inertial state estimation framework, OKVIS2, to solve the dual task of estimating camera and human states simultaneously. Our system improves the accuracy of both human and camera state estimation with respect to baseline methods by 26% and 12%, respectively, and achieves real-time performance at 15+ frames per second on an Intel i7-model CPU. Experiments were conducted on a custom dataset containing both ground truth human and camera poses collected with an indoor motion tracking system.

Xingxing Zuo, Nan Yang, Nathaniel Merrill et al.

Incrementally recovering 3D dense structures from monocular videos is of paramount importance since it enables various robotics and AR applications. Feature volumes have recently been shown to enable efficient and accurate incremental dense reconstruction without the need to first estimate depth, but they are not able to achieve as high of a resolution as depth-based methods due to the large memory consumption of high-resolution feature volumes. This letter proposes a real-time feature volume-based dense reconstruction method that predicts TSDF (Truncated Signed Distance Function) values from a novel sparsified deep feature volume, which is able to achieve higher resolutions than previous feature volume-based methods, and is favorable in large-scale outdoor scenarios where the majority of voxels are empty. An uncertainty-aware multi-view stereo (MVS) network is leveraged to infer initial voxel locations of the physical surface in a sparse feature volume. Then for refining the recovered 3D geometry, deep features are attentively aggregated from multiview images at potential surface locations, and temporally fused. Besides achieving higher resolutions than before, our method is shown to produce more complete reconstructions with finer detail in many cases. Extensive evaluations on both public and self-collected datasets demonstrate a very competitive real-time reconstruction result for our method compared to state-of-the-art reconstruction methods in both indoor and outdoor settings.

Zoe Landgraf, Raluca Scona, Tristan Laidlow et al.

By estimating 3D shape and instances from a single view, we can capture information about an environment quickly, without the need for comprehensive scanning and multi-view fusion. Solving this task for composite scenes (such as object stacks) is challenging: occluded areas are not only ambiguous in shape but also in instance segmentation; multiple decompositions could be valid. We observe that physics constrains decomposition as well as shape in occluded regions and hypothesise that a latent space learned from scenes built under physics simulation can serve as a prior to better predict shape and instances in occluded regions. To this end we propose SIMstack, a depth-conditioned Variational Auto-Encoder (VAE), trained on a dataset of objects stacked under physics simulation. We formulate instance segmentation as a centre voting task which allows for class-agnostic detection and doesn't require setting the maximum number of objects in the scene. At test time, our model can generate 3D shape and instance segmentation from a single depth view, probabilistically sampling proposals for the occluded region from the learned latent space. Our method has practical applications in providing robots some of the ability humans have to make rapid intuitive inferences of partially observed scenes. We demonstrate an application for precise (non-disruptive) object grasping of unknown objects from a single depth view.

Shuaifeng Zhi, Tristan Laidlow, Stefan Leutenegger et al.

Semantic labelling is highly correlated with geometry and radiance reconstruction, as scene entities with similar shape and appearance are more likely to come from similar classes. Recent implicit neural reconstruction techniques are appealing as they do not require prior training data, but the same fully self-supervised approach is not possible for semantics because labels are human-defined properties. We extend neural radiance fields (NeRF) to jointly encode semantics with appearance and geometry, so that complete and accurate 2D semantic labels can be achieved using a small amount of in-place annotations specific to the scene. The intrinsic multi-view consistency and smoothness of NeRF benefit semantics by enabling sparse labels to efficiently propagate. We show the benefit of this approach when labels are either sparse or very noisy in room-scale scenes. We demonstrate its advantageous properties in various interesting applications such as an efficient scene labelling tool, novel semantic view synthesis, label denoising, super-resolution, label interpolation and multi-view semantic label fusion in visual semantic mapping systems.

Marija Popovic, Florian Thomas, Sotiris Papatheodorou et al.

In robotic applications, a key requirement for safe and efficient motion planning is the ability to map obstacle-free space in unknown, cluttered 3D environments. However, commodity-grade RGB-D cameras commonly used for sensing fail to register valid depth values on shiny, glossy, bright, or distant surfaces, leading to missing data in the map. To address this issue, we propose a framework leveraging probabilistic depth completion as an additional input for spatial mapping. We introduce a deep learning architecture providing uncertainty estimates for the depth completion of RGB-D images. Our pipeline exploits the inferred missing depth values and depth uncertainty to complement raw depth images and improve the speed and quality of free space mapping. Evaluations on synthetic data show that our approach maps significantly more correct free space with relatively low error when compared against using raw data alone in different indoor environments; thereby producing more complete maps that can be directly used for robotic navigation tasks. The performance of our framework is validated using real-world data.

Yiduo Wang, Nils Funk, Milad Ramezani et al.

We present an efficient, elastic 3D LiDAR reconstruction framework which can reconstruct up to maximum LiDAR ranges (60 m) at multiple frames per second, thus enabling robot exploration in large-scale environments. Our approach only requires a CPU. We focus on three main challenges of large-scale reconstruction: integration of long-range LiDAR scans at high frequency, the capacity to deform the reconstruction after loop closures are detected, and scalability for long-duration exploration. Our system extends upon a state-of-the-art efficient RGB-D volumetric reconstruction technique, called supereight, to support LiDAR scans and a newly developed submapping technique to allow for dynamic correction of the 3D reconstruction. We then introduce a novel pose graph clustering and submap fusion feature to make the proposed system more scalable for large environments. We evaluate the performance using two public datasets including outdoor exploration with a handheld device and a drone, and with a mobile robot exploring an underground room network. Experimental results demonstrate that our system can reconstruct at 3 Hz with 60 m sensor range and ~5 cm resolution, while state-of-the-art approaches can only reconstruct to 25 cm resolution or 20 m range at the same frequency.

Nils Funk, Juan Tarrio, Sotiris Papatheodorou et al.

With the aim of bridging the gap between high quality reconstruction and mobile robot motion planning, we propose an efficient system that leverages the concept of adaptive-resolution volumetric mapping, which naturally integrates with the hierarchical decomposition of space in an octree data structure. Instead of a Truncated Signed Distance Function (TSDF), we adopt mapping of occupancy probabilities in log-odds representation, which allows to represent both surfaces, as well as the entire free, i.e. observed space, as opposed to unobserved space. We introduce a method for choosing resolution -- on the fly -- in real-time by means of a multi-scale max-min pooling of the input depth image. The notion of explicit free space mapping paired with the spatial hierarchy in the data structure, as well as map resolution, allows for collision queries, as needed for robot motion planning, at unprecedented speed. We quantitatively evaluate mapping accuracy, memory, runtime performance, and planning performance showing improvements over the state of the art, particularly in cases requiring high resolution maps.

Binbin Xu, Andrew J. Davison, Stefan Leutenegger

Dense image alignment from RGB-D images remains a critical issue for real-world applications, especially under challenging lighting conditions and in a wide baseline setting. In this paper, we propose a new framework to learn a pixel-wise deep feature map and a deep feature-metric uncertainty map predicted by a Convolutional Neural Network (CNN), which together formulate a deep probabilistic feature-metric residual of the two-view constraint that can be minimised using Gauss-Newton in a coarse-to-fine optimisation framework. Furthermore, our network predicts a deep initial pose for faster and more reliable convergence. The optimisation steps are differentiable and unrolled to train in an end-to-end fashion. Due to its probabilistic essence, our approach can easily couple with other residuals, where we show a combination with ICP. Experimental results demonstrate state-of-the-art performances on the TUM RGB-D dataset and the 3D rigid object tracking dataset. We further demonstrate our method's robustness and convergence qualitatively.

Dimos Tzoumanikas, Felix Graule, Qingyue Yan et al.

Aerial manipulation aims at combining the manoeuvrability of aerial vehicles with the manipulation capabilities of robotic arms. This, however, comes at the cost of the additional control complexity due to the coupling of the dynamics of the two systems. In this paper we present a NMPC specifically designed for MAVs equipped with a robotic arm. We formulate a hybrid control model for the combined MAV-arm system which incorporates interaction forces acting on the end effector. We explain the practical implementation of our algorithm and show extensive experimental results of our custom built system performing multiple aerial-writing tasks on a whiteboard, revealing accuracy in the order of millimetres.

Joseph Ortiz, Mark Pupilli, Stefan Leutenegger et al.

Graph processors such as Graphcore's Intelligence Processing Unit (IPU) are part of the major new wave of novel computer architecture for AI, and have a general design with massively parallel computation, distributed on-chip memory and very high inter-core communication bandwidth which allows breakthrough performance for message passing algorithms on arbitrary graphs. We show for the first time that the classical computer vision problem of bundle adjustment (BA) can be solved extremely fast on a graph processor using Gaussian Belief Propagation. Our simple but fully parallel implementation uses the 1216 cores on a single IPU chip to, for instance, solve a real BA problem with 125 keyframes and 1919 points in under 40ms, compared to 1450ms for the Ceres CPU library. Further code optimisation will surely increase this difference on static problems, but we argue that the real promise of graph processing is for flexible in-place optimisation of general, dynamically changing factor graphs representing Spatial AI problems. We give indications of this with experiments showing the ability of GBP to efficiently solve incremental SLAM problems, and deal with robust cost functions and different types of factors.

Zoe Landgraf, Fabian Falck, Michael Bloesch et al.

Generally capable Spatial AI systems must build persistent scene representations where geometric models are combined with meaningful semantic labels. The many approaches to labelling scenes can be divided into two clear groups: view-based which estimate labels from the input view-wise data and then incrementally fuse them into the scene model as it is built; and map-based which label the generated scene model. However, there has so far been no attempt to quantitatively compare view-based and map-based labelling. Here, we present an experimental framework and comparison which uses real-time height map fusion as an accessible platform for a fair comparison, opening up the route to further systematic research in this area.

Ujwal Bonde, Pablo F. Alcantarilla, Stefan Leutenegger

In this work we introduce a new Bounding-Box Free Network (BBFNet) for panoptic segmentation. Panoptic segmentation is an ideal problem for proposal-free methods as it already requires per-pixel semantic class labels. We use this observation to exploit class boundaries from off-the-shelf semantic segmentation networks and refine them to predict instance labels. Towards this goal BBFNet predicts coarse watershed levels and uses them to detect large instance candidates where boundaries are well defined. For smaller instances, whose boundaries are less reliable, BBFNet also predicts instance centers by means of Hough voting followed by mean-shift to reliably detect small objects. A novel triplet loss network helps merging fragmented instances while refining boundary pixels. Our approach is distinct from previous works in panoptic segmentation that rely on a combination of a semantic segmentation network with a computationally costly instance segmentation network based on bounding box proposals, such as Mask R-CNN, to guide the prediction of instance labels using a Mixture-of-Expert (MoE) approach. We benchmark our proposal-free method on Cityscapes and Microsoft COCO datasets and show competitive performance with other MoE based approaches while outperforming existing non-proposal based methods on the COCO dataset. We show the flexibility of our method using different semantic segmentation backbones.

Dimos Tzoumanikas, Qingyue Yan, Stefan Leutenegger

Safe and precise reference tracking is a crucial characteristic of MAVs that have to operate under the influence of external disturbances in cluttered environments. In this paper, we present a NMPC that exploits the fully physics based non-linear dynamics of the system. We furthermore show how the moment and thrust control inputs can be transformed into feasible actuator commands. In order to guarantee safe operation despite potential loss of a motor under which we show our system keeps operating safely, we developed an EKF based motor failure identification algorithm. We verify the effectiveness of the developed pipeline in flight experiments with and without motor failures.

Anna Dai, Sotiris Papatheodorou, Nils Funk et al.

Exploration and collision-free navigation through an unknown environment is a fundamental task for autonomous robots. In this paper, a novel exploration strategy for Micro Aerial Vehicles (MAVs) is presented. The goal of the exploration strategy is the reduction of map entropy regarding occupancy probabilities, which is reflected in a utility function to be maximised. We achieve fast and efficient exploration performance with tight integration between our octree-based occupancy mapping approach, frontier extraction, and motion planning-as a hybrid between frontier-based and sampling-based exploration methods. The computationally expensive frontier clustering employed in classic frontier-based exploration is avoided by exploiting the implicit grouping of frontier voxels in the underlying octree map representation. Candidate next-views are sampled from the map frontiers and are evaluated using a utility function combining map entropy and travel time, where the former is computed efficiently using sparse raycasting. These optimisations along with the targeted exploration of frontier-based methods result in a fast and computationally efficient exploration planner. The proposed method is evaluated using both simulated and real-world experiments, demonstrating clear advantages over state-of-the-art approaches.

Guillermo Gallego, Tobi Delbruck, Garrick Orchard et al.

Event cameras are bio-inspired sensors that differ from conventional frame cameras: Instead of capturing images at a fixed rate, they asynchronously measure per-pixel brightness changes, and output a stream of events that encode the time, location and sign of the brightness changes. Event cameras offer attractive properties compared to traditional cameras: high temporal resolution (in the order of microseconds), very high dynamic range (140 dB vs. 60 dB), low power consumption, and high pixel bandwidth (on the order of kHz) resulting in reduced motion blur. Hence, event cameras have a large potential for robotics and computer vision in challenging scenarios for traditional cameras, such as low-latency, high speed, and high dynamic range. However, novel methods are required to process the unconventional output of these sensors in order to unlock their potential. This paper provides a comprehensive overview of the emerging field of event-based vision, with a focus on the applications and the algorithms developed to unlock the outstanding properties of event cameras. We present event cameras from their working principle, the actual sensors that are available and the tasks that they have been used for, from low-level vision (feature detection and tracking, optic flow, etc.) to high-level vision (reconstruction, segmentation, recognition). We also discuss the techniques developed to process events, including learning-based techniques, as well as specialized processors for these novel sensors, such as spiking neural networks. Additionally, we highlight the challenges that remain to be tackled and the opportunities that lie ahead in the search for a more efficient, bio-inspired way for machines to perceive and interact with the world.

Shuaifeng Zhi, Michael Bloesch, Stefan Leutenegger et al.

Systems which incrementally create 3D semantic maps from image sequences must store and update representations of both geometry and semantic entities. However, while there has been much work on the correct formulation for geometrical estimation, state-of-the-art systems usually rely on simple semantic representations which store and update independent label estimates for each surface element (depth pixels, surfels, or voxels). Spatial correlation is discarded, and fused label maps are incoherent and noisy. We introduce a new compact and optimisable semantic representation by training a variational auto-encoder that is conditioned on a colour image. Using this learned latent space, we can tackle semantic label fusion by jointly optimising the low-dimenional codes associated with each of a set of overlapping images, producing consistent fused label maps which preserve spatial correlation. We also show how this approach can be used within a monocular keyframe based semantic mapping system where a similar code approach is used for geometry. The probabilistic formulation allows a flexible formulation where we can jointly estimate motion, geometry and semantics in a unified optimisation.

Andrea Nicastro, Ronald Clark, Stefan Leutenegger

Detailed 3D reconstruction is an important challenge with application to robotics, augmented and virtual reality, which has seen impressive progress throughout the past years. Advancements were driven by the availability of depth cameras (RGB-D), as well as increased compute power, e.g.\ in the form of GPUs -- but also thanks to inclusion of machine learning in the process. Here, we propose X-Section, an RGB-D 3D reconstruction approach that leverages deep learning to make object-level predictions about thicknesses that can be readily integrated into a volumetric multi-view fusion process, where we propose an extension to the popular KinectFusion approach. In essence, our method allows to complete shape in general indoor scenes behind what is sensed by the RGB-D camera, which may be crucial e.g.\ for robotic manipulation tasks or efficient scene exploration. Predicting object thicknesses rather than volumes allows us to work with comparably high spatial resolution without exploding memory and training data requirements on the employed Convolutional Neural Networks. In a series of qualitative and quantitative evaluations, we demonstrate how we accurately predict object thickness and reconstruct general 3D scenes containing multiple objects.