With the recent rise of large language models, vision-language models, and other general foundation models, there is growing potential for multimodal, multi-task robotics that can operate in diverse environments given natural language input. One such application is indoor navigation using natural language instructions. However, despite recent progress, this problem remains challenging due to the 3D spatial reasoning and semantic understanding required. Additionally, the language used may be imperfect or misaligned with the scene, further complicating the task. To address this challenge, we curate a benchmark dataset, IRef-VLA, for Interactive Referential Vision and Language-guided Action in 3D Scenes with imperfect references. IRef-VLA is the largest real-world dataset for the referential grounding task, consisting of over 11.5 K scanned 3D rooms from existing datasets, 7.6 M heuristically generated semantic relations, and 4.7 M referential statements. Our dataset also contains semantic object and room annotations, scene graphs, navigable free space annotations, and is augmented with statements where the language has imperfections or ambiguities. We verify the generalizability of our dataset by evaluating with state-of-the-art models to obtain a performance baseline and also develop a graphsearch baseline to demonstrate the performance bound and generation of alternatives using scene-graph knowledge. With this benchmark, we aim to provide a resource for 3D scene understanding that aids the development of robust, interactive navigation systems. The dataset and all source code is publicly released11https://github.com/HaochenZ11/IRef-VLA.

Interpreting object-referential language and grounding objects in 3D with spatial relations and attributes is essential for robots operating alongside humans. However, this task is often challenging due to the diversity of scenes, large number of fine-grained objects, and complex free-form nature of language references. Furthermore, in the 3D domain, obtaining large amounts of natural language training data is difficult. Thus, it is important for methods to learn from little data and zero-shot generalize to new environments. To address these challenges, we propose SORT3D, an approach that utilizes rich object attributes from 2D data and merges a heuristics-based spatial reasoning toolbox with the ability of large language models (LLMs) to perform sequential reasoning. Importantly, our method does not require text-to-3D data for training and can be applied zero-shot to unseen environments. We show that SORT3D achieves state-of-the-art zero-shot performance on complex view-dependent grounding tasks on two benchmarks. We also implement the pipeline to run real-time on two autonomous vehicles and demonstrate that our approach can be used for object-goal navigation on previously unseen real-world environments. All source code for the system pipeline is publicly released.1

Estimating agent pose and 3D scene structure from multi-camera rigs is a central task in embodied AI applications such as autonomous driving. Recent learned approaches such as DUSt3R have shown impressive performance in multiview settings. However, these models treat images as unstructured collections, limiting effectiveness in scenarios where frames are captured from synchronized rigs with known or inferable structure. To this end, we introduce Rig3R, a generalization of prior multiview reconstruction models that incorporates rig structure when available, and learns to infer it when not. Rig3R conditions on optional rig metadata including camera ID, time, and rig poses to develop a rig-aware latent space that remains robust to missing information. It jointly predicts pointmaps and two types of raymaps: a pose raymap relative to a global frame, and a rig raymap relative to a rig-centric frame consistent across time. Rig raymaps allow the model to infer rig structure directly from input images when metadata is missing. The global pose raymaps allow the model to reason about the agent’s ego-motion, while the rig raymaps allow the model to infer rig structure directly from input images when metadata is missing. Rig3R achieves state-of-the-art performance in 3D reconstruction, camera pose estimation, and rig discovery -- outperforming both traditional and learned methods by 17-45% mAA across diverse real-world rig datasets, all in a single forward pass without post-processing or iterative refinement.

Localization is a computer vision task by which the position and orientation of a camera is determined from an image and environmental map. We propose a method for performing localization in a privacy preserving manner supporting two scenarios: first, when the image and map are held by a client who wishes to offload localization to untrusted third parties, and second, when the image and map are held separately by untrusting parties. Privacy preserving localization is necessary when the image and map are confidential, and offloading conserves on-device power and frees resources for other tasks. To accomplish this we integrate existing localization methods and secure multi-party computation (MPC), specifically garbled circuits, yielding proof-based security guarantees in contrast to existing obfuscation-based approaches which recent related work has shown vulnerable. We present two approaches to localization, a baseline data-oblivious adaptation of localization suitable for garbled circuits and our novel Single Iteration Localization. Our technique improves overall performance while maintaining confidentiality of the input image, map, and output pose at the expense of increased communication rounds but reduced computation and communication required per round. Single Iteration Localization is over two orders of magnitude faster than a straightforward application of garbled circuits to localization enabling real-world usage in Turbo the Snail, the first robot to offload localization without revealing input images, environmental map, position, or orientation to offload servers.

Localization is a common task in computer vision by which the position and orientation of a camera is determined from an image and 3D map of the environment. We propose a demonstration of securely performing localization in a privacy preserving manner. This allows localization so that a lightweight client device, such as a mobile robot, may offload the computation to an untrusted server without revealing anything about their data. To accomplish this goal, we combine existing localization methods with secure multiparty computation (MPC), specifically garbled circuits. As such, the security guarantees of this work are simulation-based in contrast to existing obfuscation-based approaches to pose estimation for which privacy is inversely proportional to input size. We propose two approaches, a baseline data-oblivious adaptation of localization suitable for MPC and Single Iteration Localization which runs each localization step individually, improving performance at the cost of round complexity while maintaining confidentiality of the input image, map, and output pose. Single Iteration Localization is over two orders of magnitude faster than the data-oblivious approach enabling real-world usage in Turbo the Snail, the first robot to offload localization without revealing input images, environmental map, position, or orientation to offload servers.