Robotic grasping is a cornerstone capability of embodied systems. Many methods directly output grasps from partial information without modeling the geometry of the scene, leading to suboptimal motion and even collisions. To address these issues, we introduce ZeroGrasp, a novel framework that simultaneously performs 3D reconstruction and grasp pose prediction in near real-time. A key insight of our method is that occlusion reasoning and modeling the spatial relationships between objects is beneficial for both accurate reconstruction and grasping. We couple our method with a novel large-scale synthetic dataset, which comprises 1M photo-realistic images, high-resolution 3D reconstructions and 11.3B physically-valid grasp pose annotations for 12K objects from the Objaverse-LVIS dataset. We evaluate Zero-Grasp on the GraspNet-1B benchmark as well as through real-world robot experiments. ZeroGrasp achieves state-of-the-art performance and generalizes to novel real-world objects by leveraging synthetic data. https://sh8.io/#/zerograsp

We would like to estimate the pose and full shape of an object from a single observation, without assuming known 3D model or category. In this work, we propose OmniShape, the first method of its kind to enable probabilistic pose and shape estimation. OmniShape is based on the key insight that shape completion can be decoupled into two multi-modal distributions: one capturing how measurements project into a normalized object reference frame defined by the dataset and the other modelling a prior over object geometries represented as triplanar neural fields. By training separate conditional diffusion models for these two distributions, we enable sampling multiple hypotheses from the joint pose and shape distri-bution. OmniShape demonstrates compelling performance on challenging real world datasets. Project website: https://tri-ml.glthub.io/omnishape.

This paper addresses the challenging problem of category-level pose estimation. Current state-of-the-art methods for this task face challenges when dealing with symmetric objects and when attempting to generalize to new environments solely through synthetic data training. In this work, we address these challenges by proposing a probabilistic model that relies on diffusion to estimate dense canonical maps crucial for recovering partial object shapes as well as establishing correspondences essential for pose estimation. Furthermore, we introduce critical components to enhance performance by leveraging the strength of the diffusion models with multi-modal input representations. We demonstrate the effectiveness of our method by testing it on a range of real datasets. Despite being trained solely on our generated synthetic data, our approach achieves state-of-the-art performance and unprecedented generalization qualities, outperforming baselines, even those specifically trained on the target domain. Our code and data for the generalization benchmark can be found at https://woven-planet.github.io/DiffusionNOCS/.

Incorporating inductive bias by embedding geometric entities (such as rays) as input has proven successful in multi-view learning. However, the methods adopting this technique typically lack equivariance, which is crucial for effective 3D learning. Equivariance serves as a valuable inductive prior, aiding in the generation of robust multi-view features for 3D scene understanding. In this paper, we explore the application of equivariant multi-view learning to depth estimation, not only recognizing its significance for computer vision and robotics but also addressing the limitations of previous research. Most prior studies have either overlooked equivariance in this setting or achieved only approximate equivariance through data augmentation, which often leads to inconsistencies across different reference frames. To address this issue, we propose to embed $SE(3)$ equivariance into the Perceiver IO architecture. We employ Spherical Harmonics for positional encoding to ensure 3D rotation equivariance, and develop a specialized equivariant encoder and decoder within the Perceiver IO architecture. To validate our model, we applied it to the task of stereo depth estimation, achieving state of the art results on real-world datasets without explicit geometric constraints or extensive data augmentation.

We present CARTO, a novel approach for reconstructing multiple articulated objects from a single stereo RGB observation. We use implicit object-centric representations and learn a single geometry and articulation decoder for multiple object categories. Despite training on multiple categories, our decoder achieves a comparable reconstruction accuracy to methods that train bespoke decoders separately for each category. Combined with our stereo image encoder we infer the 3D shape, 6D pose, size, joint type, and the joint state of multiple unknown objects in a single forward pass. Our method achieves a 20.4% absolute improvement in mAP 3D IOU50 for novel instances when compared to a two-stage pipeline. Inference time is fast and can run on a NVIDIA TITAN XP GPU at 1 HZ for eight or less objects present. While only trained on simulated data, CARTO transfers to real-world object instances. Code and evaluation data is available at: carto. cs. uni - freiburg. de

Learned visual dynamics models have proven effective for robotic manipulation tasks. Yet, it remains unclear how best to represent scenes involving multi-object interactions. Current methods decompose a scene into discrete objects, but they struggle with precise modeling and manipulation amid challenging lighting conditions as they only encode appearance tied with specific illuminations. In this work, we propose using object-centric neural scattering functions (OSFs) as object representations in a model-predictive control framework. OSFs model per-object light transport, enabling compositional scene re-rendering under object rearrangement and varying lighting conditions. By combining this approach with inverse parameter estimation and graph-based neural dynamics models, we demonstrate improved model-predictive control performance and generalization in compositional multi-object environments, even in previously unseen scenarios and harsh lighting conditions.

Recent implicit neural representations have shown great results for novel view synthesis. However, existing methods require expensive per-scene optimization from many views hence limiting their application to real-world unbounded urban settings where the objects of interest or backgrounds are observed from very few views. To mitigate this challenge, we introduce a new approach called NeO 360, Neural fields for sparse view synthesis of outdoor scenes. NeO 360 is a generalizable method that reconstructs 360° scenes from a single or a few posed RGB images. The essence of our approach is in capturing the distribution of complex real-world outdoor 3D scenes and using a hybrid image-conditional triplanar representation that can be queried from any world point. Our representation combines the best of both voxel-based and bird’s-eye-view (BEV) representations and is more effective and expressive than each. NeO 360’s representation allows us to learn from a large collection of unbounded 3D scenes while offering generalizability to new views and novel scenes from as few as a single image during inference. We demonstrate our approach on the pro posed challenging 360° unbounded dataset, called NeRDS 360, and show that NeO 360 outperforms state-of-the-art generalizable methods for novel view synthesis while also offering editing and composition capabilities. Project page: zubair-irshad.github.io/projects/neo360.html

We propose a hierarchical representation of objects, where the representation of each object is allowed to change based on the quality of accumulated measurements. We initially estimate each object as a 2D bounding box or a 3D point, encoding only the geometric properties that can be well-constrained using limited viewpoints. With additional measurements, we allow each object to become a higher dimensional 3D volumetric model for improved reconstruction accuracy and collision-testing. Our Hierarchical Object Map Estimation (HOME) is robust to deficiencies in viewpoints and allows planning safe and efficient trajectories around object obstacles using a monocular camera. We demonstrate the advantages of our approach on a real-world TUM dataset and during visual-inertial navigation of a quad-rotor in simulation.

Deep learning approaches to estimating 3D object pose and geometry present an attractive alternative to online estimation techniques, which can suffer from significant estimation latency. However, a practical hurdle to training state-of-the-art deep 3D bounding box estimators is collecting a sufficiently large dataset of 3D bounding box labels. In this work, we present a novel framework for weakly supervised volumetric monocular estimation (VoluMon) that requires annotations in the image space only, i.e., associated object bounding box detections and instance segmentation. By approximating object geometry as ellipsoids, we can exploit the dual form of the ellipsoid to optimize with respect to bounding box annotations and the primal form of the ellipsoid to optimize with respect to a segmented pointcloud. For a simulated dataset with access to ground-truth, we show monocular object estimation performance similar to a naive online depth based estimation approach and after online refinement when depth images are available, we also approach the performance of a learned deep 6D pose estimator, which is supervised with projected 3D bounding box keypoints and assumes known model dimensions. Finally, we show promising qualitative results generated from a real-world dataset collected using a stereo pair.

We would like to enable a robot to navigate efficiently and robustly in known, structured environments that are large enough to cause traditional planning approaches to incur considerable computational cost. Hierarchical planners are a promising way to increase planning efficiency in such environments because high-level abstract plans can be used to reduce the size of the search space over which detailed planning occurs. However, useful high-level representations of planning problems can be challenging to generate without prior domain knowledge. In this work, we propose a high-level planning representation which can be learned from previous plans considered in the environment and used online during hierarchical, multi-query robot navigation. We treat previous planning results as noisy measurements of high-level navigation properties, then update these properties over time using recursive estimation. We test our approach in standard and risk-aware hierarchical planning schemes, and demonstrate up to an 86% decrease in the number of nodes expanded and a 66% decrease in wallclock time as compared to a baseline A* planner while finding plans that are only 2-10% more expensive.

We would like to enable a robotic agent to quickly and intelligently find promising trajectories through structured, unknown environments. Many approaches to navigation in unknown environments are limited to considering geometric information only, which leads to myopic behavior. In this work, we show that learning a sampling distribution that incorporates both geometric information and explicit, object-level semantics for sampling-based planners enables efficient planning at longer horizons in partially-known environments. We demonstrate that our learned planner is up to 2.7 times more likely to find a plan than the baseline, and can result in up to a 16% reduction in traversal costs as calculated by linear regression. We also show promising qualitative results on real-world data.

We present Robust Object-based SLAM for High-speed Autonomous Navigation (ROSHAN), a novel approach to object-level mapping suitable for autonomous navigation. In ROSHAN, we represent objects as ellipsoids and infer their parameters using three sources of information - bounding box detections, image texture, and semantic knowledge - to overcome the observability problem in ellipsoid-based SLAM under common forward-translating vehicle motions. Each bounding box provides four planar constraints on an object surface and we add a fifth planar constraint using the texture on the objects along with a semantic prior on the shape of ellipsoids. We demonstrate ROSHAN in simulation where we outperform the baseline, reducing the median shape error by 83% and the median position error by 72% in a forward-moving camera sequence. We demonstrate similar qualitative result on data collected on a fast-moving autonomous quadrotor.

We present a novel method of measurement covariance estimation that models measurement uncertainty as a function of the measurement itself. Existing work in predictive sensor modeling outperforms conventional fixed models, but requires domain knowledge of the sensors that heavily influences the accuracy and the computational cost of the models. In this work, we introduce Deep Inference for Covariance Estimation (DICE), which utilizes a deep neural network to predict the covariance of a sensor measurement from raw sensor data. We show that given pairs of raw sensor measurement and ground-truth measurement error, we can learn a representation of the measurement model via supervised regression on the prediction performance of the model, eliminating the need for hand-coded features and parametric forms. Our approach is sensor-agnostic, and we demonstrate improved covariance prediction on both simulated and real data.