Zhaoshuo Li

Niket Agarwal, Arslan Ali, Maciej Bala et al. · nvidia

Physical AI needs to be trained digitally first. It needs a digital twin of itself, the policy model, and a digital twin of the world, the world model. In this paper, we present the Cosmos World Foundation Model Platform to help developers build customized world models for their Physical AI setups. We position a world foundation model as a general-purpose world model that can be fine-tuned into customized world models for downstream applications. Our platform covers a video curation pipeline, pre-trained world foundation models, examples of post-training of pre-trained world foundation models, and video tokenizers. To help Physical AI builders solve the most critical problems of our society, we make Cosmos open-source and our models open-weight with permissive licenses available via https://github.com/nvidia-cosmos/cosmos-predict1.

Alisson Azzolini, Junjie Bai, Hannah Brandon et al. · nvidia

Physical AI systems need to perceive, understand, and perform complex actions in the physical world. In this paper, we present the Cosmos-Reason1 models that can understand the physical world and generate appropriate embodied decisions (e.g., next step action) in natural language through long chain-of-thought reasoning processes. We begin by defining key capabilities for Physical AI reasoning, with a focus on physical common sense and embodied reasoning. To represent physical common sense, we use a hierarchical ontology that captures fundamental knowledge about space, time, and physics. For embodied reasoning, we rely on a two-dimensional ontology that generalizes across different physical embodiments. Building on these capabilities, we develop two multimodal large language models, Cosmos-Reason1-7B and Cosmos-Reason1-56B. We curate data and train our models in two stages: Physical AI supervised fine-tuning (SFT) and Physical AI reinforcement learning (RL). To evaluate our models, we build comprehensive benchmarks for physical common sense and embodied reasoning according to our ontologies. Evaluation results show that Physical AI SFT and RL bring significant improvements. To facilitate the development of Physical AI, we make our code and pre-trained models available under the NVIDIA Open Model License at https://github.com/nvidia-cosmos/cosmos-reason1.

Arslan Ali, Junjie Bai, Maciej Bala et al. · nvidia

We introduce [Cosmos-Predict2.5], the latest generation of the Cosmos World Foundation Models for Physical AI. Built on a flow-based architecture, [Cosmos-Predict2.5] unifies Text2World, Image2World, and Video2World generation in a single model and leverages [Cosmos-Reason1], a Physical AI vision-language model, to provide richer text grounding and finer control of world simulation. Trained on 200M curated video clips and refined with reinforcement learning-based post-training, [Cosmos-Predict2.5] achieves substantial improvements over [Cosmos-Predict1] in video quality and instruction alignment, with models released at 2B and 14B scales. These capabilities enable more reliable synthetic data generation, policy evaluation, and closed-loop simulation for robotics and autonomous systems. We further extend the family with [Cosmos-Transfer2.5], a control-net style framework for Sim2Real and Real2Real world translation. Despite being 3.5$\times$ smaller than [Cosmos-Transfer1], it delivers higher fidelity and robust long-horizon video generation. Together, these advances establish [Cosmos-Predict2.5] and [Cosmos-Transfer2.5] as versatile tools for scaling embodied intelligence. To accelerate research and deployment in Physical AI, we release source code, pretrained checkpoints, and curated benchmarks under the NVIDIA Open Model License at https://github.com/nvidia-cosmos/cosmos-predict2.5 and https://github.com/nvidia-cosmos/cosmos-transfer2.5. We hope these open resources lower the barrier to adoption and foster innovation in building the next generation of embodied intelligence.

Aditi, Niket Agarwal, Arslan Ali et al.

We introduce Cosmos 3, a family of omnimodal world models designed to jointly process and generate language, image, video, audio, and action sequences within a unified mixture-of-transformers architecture. By supporting highly flexible input-output configurations, Cosmos 3 seamlessly unifies critical modalities for Physical AI -- effectively subsuming vision-language models, video generators, world simulators, and world-action models into a single framework. Our evaluation demonstrates that Cosmos 3 establishes a new state-of-the-art across a diverse suite of understanding and generation tasks, demonstrating omnimodal world models as scalable, general-purpose backbones for embodied agents. Our post-trained Cosmos 3 models were ranked as the best open-source Text-to-Image and Image-to-Video models by Artificial Analysis, and the best policy model by RoboArena at the time the technical report was written. To accelerate open research and deployment in Physical AI, we make our code, model checkpoints, curated synthetic datasets, and evaluation benchmark available under the Linux Foundation's OpenMDW-1.1 https://openmdw.ai/license/1-1/ License at https://github.com/nvidia/cosmos}{github.com/nvidia/cosmos and https://huggingface.co/collections/nvidia/cosmos3 . The project website is available at https://research.nvidia.com/labs/cosmos-lab/cosmos3 .

Maciej Bala, Yin Cui, Yifan Ding et al. · nvidia

We introduce Edify 3D, an advanced solution designed for high-quality 3D asset generation. Our method first synthesizes RGB and surface normal images of the described object at multiple viewpoints using a diffusion model. The multi-view observations are then used to reconstruct the shape, texture, and PBR materials of the object. Our method can generate high-quality 3D assets with detailed geometry, clean shape topologies, high-resolution textures, and materials within 2 minutes of runtime.

Weiyu Guo, Zhaoshuo Li, Yongkui Yang et al.

Stereo depth estimation is of great interest for computer vision research. However, existing methods struggles to generalize and predict reliably in hazardous regions, such as large uniform regions. To overcome these limitations, we propose Context Enhanced Path (CEP). CEP improves the generalization and robustness against common failure cases in existing solutions by capturing the long-range global information. We construct our stereo depth estimation model, Context Enhanced Stereo Transformer (CSTR), by plugging CEP into the state-of-the-art stereo depth estimation method Stereo Transformer. CSTR is examined on distinct public datasets, such as Scene Flow, Middlebury-2014, KITTI-2015, and MPI-Sintel. We find CSTR outperforms prior approaches by a large margin. For example, in the zero-shot synthetic-to-real setting, CSTR outperforms the best competing approaches on Middlebury-2014 dataset by 11%. Our extensive experiments demonstrate that the long-range information is critical for stereo matching task and CEP successfully captures such information.

Zhaoshuo Li, Thomas Müller, Alex Evans et al.

Neural surface reconstruction has been shown to be powerful for recovering dense 3D surfaces via image-based neural rendering. However, current methods struggle to recover detailed structures of real-world scenes. To address the issue, we present Neuralangelo, which combines the representation power of multi-resolution 3D hash grids with neural surface rendering. Two key ingredients enable our approach: (1) numerical gradients for computing higher-order derivatives as a smoothing operation and (2) coarse-to-fine optimization on the hash grids controlling different levels of details. Even without auxiliary inputs such as depth, Neuralangelo can effectively recover dense 3D surface structures from multi-view images with fidelity significantly surpassing previous methods, enabling detailed large-scale scene reconstruction from RGB video captures.

Tianchang Shen, Zhaoshuo Li, Marc Law et al. · nvidia, utoronto

Meshes are ubiquitous in visual computing and simulation, yet most existing machine learning techniques represent meshes only indirectly, e.g. as the level set of a scalar field or deformation of a template, or as a disordered triangle soup lacking local structure. This work presents a scheme to directly generate manifold, polygonal meshes of complex connectivity as the output of a neural network. Our key innovation is to define a continuous latent connectivity space at each mesh vertex, which implies the discrete mesh. In particular, our vertex embeddings generate cyclic neighbor relationships in a halfedge mesh representation, which gives a guarantee of edge-manifoldness and the ability to represent general polygonal meshes. This representation is well-suited to machine learning and stochastic optimization, without restriction on connectivity or topology. We first explore the basic properties of this representation, then use it to fit distributions of meshes from large datasets. The resulting models generate diverse meshes with tessellation structure learned from the dataset population, with concise details and high-quality mesh elements. In applications, this approach not only yields high-quality outputs from generative models, but also enables directly learning challenging geometry processing tasks such as mesh repair.

Zhaoshuo Li, Hongchao Shu, Ruixing Liang et al.

Purpose: Tracking the 3D motion of the surgical tool and the patient anatomy is a fundamental requirement for computer-assisted skull-base surgery. The estimated motion can be used both for intra-operative guidance and for downstream skill analysis. Recovering such motion solely from surgical videos is desirable, as it is compliant with current clinical workflows and instrumentation. Methods: We present Tracker of Anatomy and Tool (TAToo). TAToo jointly tracks the rigid 3D motion of patient skull and surgical drill from stereo microscopic videos. TAToo estimates motion via an iterative optimization process in an end-to-end differentiable form. For robust tracking performance, TAToo adopts a probabilistic formulation and enforces geometric constraints on the object level. Results: We validate TAToo on both simulation data, where ground truth motion is available, as well as on anthropomorphic phantom data, where optical tracking provides a strong baseline. We report sub-millimeter and millimeter inter-frame tracking accuracy for skull and drill, respectively, with rotation errors below 1°. We further illustrate how TAToo may be used in a surgical navigation setting. Conclusion: We present TAToo, which simultaneously tracks the surgical tool and the patient anatomy in skull-base surgery. TAToo directly predicts the motion from surgical videos, without the need of any markers. Our results show that the performance of TAToo compares favorably to competing approaches. Future work will include fine-tuning of our depth network to reach a 1 mm clinical accuracy goal desired for surgical applications in the skull base.

Jiaxiang Tang, Zhaoshuo Li, Zekun Hao et al.

Current auto-regressive mesh generation methods suffer from issues such as incompleteness, insufficient detail, and poor generalization. In this paper, we propose an Auto-regressive Auto-encoder (ArAE) model capable of generating high-quality 3D meshes with up to 4,000 faces at a spatial resolution of $512^3$. We introduce a novel mesh tokenization algorithm that efficiently compresses triangular meshes into 1D token sequences, significantly enhancing training efficiency. Furthermore, our model compresses variable-length triangular meshes into a fixed-length latent space, enabling training latent diffusion models for better generalization. Extensive experiments demonstrate the superior quality, diversity, and generalization capabilities of our model in both point cloud and image-conditioned mesh generation tasks.

Hao Ding, Jie Ying Wu, Zhaoshuo Li et al.

Purpose: Vision-based robot tool segmentation plays a fundamental role in surgical robots and downstream tasks. CaRTS, based on a complementary causal model, has shown promising performance in unseen counterfactual surgical environments in the presence of smoke, blood, etc. However, CaRTS requires over 30 iterations of optimization to converge for a single image due to limited observability. Method: To address the above limitations, we take temporal relation into consideration and propose a temporal causal model for robot tool segmentation on video sequences. We design an architecture named Temporally Constrained CaRTS (TC-CaRTS). TC-CaRTS has three novel modules to complement CaRTS - temporal optimization pipeline, kinematics correction network, and spatial-temporal regularization. Results: Experiment results show that TC-CaRTS requires much fewer iterations to achieve the same or better performance as CaRTS. TC- CaRTS also has the same or better performance in different domains compared to CaRTS. All three modules are proven to be effective. Conclusion: We propose TC-CaRTS, which takes advantage of temporal constraints as additional observability. We show that TC-CaRTS outperforms prior work in the robot tool segmentation task with improved convergence speed on test datasets from different domains.

Yufan He, Pengfei Guo, Mengya Xu et al.

Data scarcity remains a fundamental barrier to achieving fully autonomous surgical robots. While large scale vision language action (VLA) models have shown impressive generalization in household and industrial manipulation by leveraging paired video action data from diverse domains, surgical robotics suffers from the paucity of datasets that include both visual observations and accurate robot kinematics. In contrast, vast corpora of surgical videos exist, but they lack corresponding action labels, preventing direct application of imitation learning or VLA training. In this work, we aim to alleviate this problem by learning policy models from SurgWorld, a world model designed for surgical physical AI. We curated the Surgical Action Text Alignment (SATA) dataset with detailed action description specifically for surgical robots. Then we built SurgeWorld based on the most advanced physical AI world model and SATA. It's able to generate diverse, generalizable and realistic surgery videos. We are also the first to use an inverse dynamics model to infer pseudokinematics from synthetic surgical videos, producing synthetic paired video action data. We demonstrate that a surgical VLA policy trained with these augmented data significantly outperforms models trained only on real demonstrations on a real surgical robot platform. Our approach offers a scalable path toward autonomous surgical skill acquisition by leveraging the abundance of unlabeled surgical video and generative world modeling, thus opening the door to generalizable and data efficient surgical robot policies.

Hongchao Shu, Ruixing Liang, Zhaoshuo Li et al.

Purpose: Digital twins are virtual interactive models of the real world, exhibiting identical behavior and properties. In surgical applications, computational analysis from digital twins can be used, for example, to enhance situational awareness. Methods: We present a digital twin framework for skull-base surgeries, named Twin-S, which can be integrated within various image-guided interventions seamlessly. Twin-S combines high-precision optical tracking and real-time simulation. We rely on rigorous calibration routines to ensure that the digital twin representation precisely mimics all real-world processes. Twin-S models and tracks the critical components of skull-base surgery, including the surgical tool, patient anatomy, and surgical camera. Significantly, Twin-S updates and reflects real-world drilling of the anatomical model in frame rate. Results: We extensively evaluate the accuracy of Twin-S, which achieves an average 1.39 mm error during the drilling process. We further illustrate how segmentation masks derived from the continuously updated digital twin can augment the surgical microscope view in a mixed reality setting, where bone requiring ablation is highlighted to provide surgeons additional situational awareness. Conclusion: We present Twin-S, a digital twin environment for skull-base surgery. Twin-S tracks and updates the virtual model in real-time given measurements from modern tracking technologies. Future research on complementing optical tracking with higher-precision vision-based approaches may further increase the accuracy of Twin-S.

Hongchi Xia, Xuan Li, Zhaoshuo Li et al.

Real-world data collection for embodied agents remains costly and unsafe, calling for scalable, realistic, and simulator-ready 3D environments. However, existing scene-generation systems often rely on rule-based or task-specific pipelines, yielding artifacts and physically invalid scenes. We present SAGE, an agentic framework that, given a user-specified embodied task (e.g., "pick up a bowl and place it on the table"), understands the intent and automatically generates simulation-ready environments at scale. The agent couples multiple generators for layout and object composition with critics that evaluate semantic plausibility, visual realism, and physical stability. Through iterative reasoning and adaptive tool selection, it self-refines the scenes until meeting user intent and physical validity. The resulting environments are realistic, diverse, and directly deployable in modern simulators for policy training. Policies trained purely on this data exhibit clear scaling trends and generalize to unseen objects and layouts, demonstrating the promise of simulation-driven scaling for embodied AI. Code, demos, and the SAGE-10k dataset can be found on the project page here: https://nvlabs.github.io/sage.

Jiaxin Guo, Jiangliu Wang, Zhaoshuo Li et al.

Soft tissue tracking is crucial for computer-assisted interventions. Existing approaches mainly rely on extracting discriminative features from the template and videos to recover corresponding matches. However, it is difficult to adopt these techniques in surgical scenes, where tissues are changing in shape and appearance throughout the surgery. To address this problem, we exploit optical flow to naturally capture the pixel-wise tissue deformations and adaptively correct the tracked template. Specifically, we first implement an inter-frame matching mechanism to extract a coarse region of interest based on optical flow from consecutive frames. To accommodate appearance change and alleviate drift, we then propose an adaptive-template matching method, which updates the tracked template based on the reliability of the estimates. Our approach, Ada-Tracker, enjoys both short-term dynamics modeling by capturing local deformations and long-term dynamics modeling by introducing global temporal compensation. We evaluate our approach on the public SurgT benchmark, which is generated from Hamlyn, SCARED, and Kidney boundary datasets. The experimental results show that Ada-Tracker achieves superior accuracy and performs more robustly against prior works. Code is available at https://github.com/wrld/Ada-Tracker.

Qingqing Zhao, Yao Lu, Moo Jin Kim et al.

Vision-language-action models (VLAs) have shown potential in leveraging pretrained vision-language models and diverse robot demonstrations for learning generalizable sensorimotor control. While this paradigm effectively utilizes large-scale data from both robotic and non-robotic sources, current VLAs primarily focus on direct input--output mappings, lacking the intermediate reasoning steps crucial for complex manipulation tasks. As a result, existing VLAs lack temporal planning or reasoning capabilities. In this paper, we introduce a method that incorporates explicit visual chain-of-thought (CoT) reasoning into vision-language-action models (VLAs) by predicting future image frames autoregressively as visual goals before generating a short action sequence to achieve these goals. We introduce CoT-VLA, a state-of-the-art 7B VLA that can understand and generate visual and action tokens. Our experimental results demonstrate that CoT-VLA achieves strong performance, outperforming the state-of-the-art VLA model by 17% in real-world manipulation tasks and 6% in simulation benchmarks. Project website: https://cot-vla.github.io/

Haotian Xue, Yunhao Ge, Yu Zeng et al. · gatech

Vision-Language Models (VLMs) have demonstrated impressive world knowledge across a wide range of tasks, making them promising candidates for embodied reasoning applications. However, existing benchmarks primarily evaluate the embodied reasoning ability of VLMs through multiple-choice questions based on image annotations -- for example, selecting which trajectory better describes an event in the image. In this work, we introduce the Point-It-Out (PIO) benchmark, a novel benchmark designed to systematically assess the embodied reasoning abilities of VLMs through precise visual grounding. We propose a hierarchical evaluation protocol spanning three stages (S1: referred-object localization, S2: task-driven pointing, and S3: visual trace prediction), with data collected from critical domains for embodied intelligence, including indoor, kitchen, driving, and robotic manipulation scenarios. Extensive experiments with over ten state-of-the-art VLMs reveal several interesting findings. For example, strong general-purpose models such as GPT-4o, while excelling on many benchmarks (e.g., language, perception, and reasoning), underperform compared to some open-source models in precise visual grounding; models such as MoLMO perform well in S1 and S2 but struggle in S3, where requires grounding combined with visual trace planning.

Xingtong Liu, Zhaoshuo Li, Masaru Ishii et al.

In endoscopy, many applications (e.g., surgical navigation) would benefit from a real-time method that can simultaneously track the endoscope and reconstruct the dense 3D geometry of the observed anatomy from a monocular endoscopic video. To this end, we develop a Simultaneous Localization and Mapping system by combining the learning-based appearance and optimizable geometry priors and factor graph optimization. The appearance and geometry priors are explicitly learned in an end-to-end differentiable training pipeline to master the task of pair-wise image alignment, one of the core components of the SLAM system. In our experiments, the proposed SLAM system is shown to robustly handle the challenges of texture scarceness and illumination variation that are commonly seen in endoscopy. The system generalizes well to unseen endoscopes and subjects and performs favorably compared with a state-of-the-art feature-based SLAM system. The code repository is available at https://github.com/lppllppl920/SAGE-SLAM.git.

Zhaoshuo Li, Alex Gordon, Thomas Looi et al.

This paper presents a dynamic constraint formulation to provide protective virtual fixtures of 3D anatomical structures from polygon mesh representations. The proposed approach can anisotropically limit the tool motion of surgical robots without any assumption of the local anatomical shape close to the tool. Using a bounded search strategy and Principle Directed tree, the proposed system can run efficiently at 180 Hz for a mesh object containing 989,376 triangles and 493,460 vertices. The proposed algorithm has been validated in both simulation and skull cutting experiments. The skull cutting experiment setup uses a novel piezoelectric bone cutting tool designed for the da Vinci research kit. The result shows that the virtual fixture assisted teleoperation has statistically significant improvements in the cutting path accuracy and penetration depth control. The code has been made publicly available at https://github.com/mli0603/PolygonMeshVirtualFixture.

Zhendong Wang, Zhaoshuo Li, Ajay Mandlekar et al.

Diffusion models, praised for their success in generative tasks, are increasingly being applied to robotics, demonstrating exceptional performance in behavior cloning. However, their slow generation process stemming from iterative denoising steps poses a challenge for real-time applications in resource-constrained robotics setups and dynamically changing environments. In this paper, we introduce the One-Step Diffusion Policy (OneDP), a novel approach that distills knowledge from pre-trained diffusion policies into a single-step action generator, significantly accelerating response times for robotic control tasks. We ensure the distilled generator closely aligns with the original policy distribution by minimizing the Kullback-Leibler (KL) divergence along the diffusion chain, requiring only $2\%$-$10\%$ additional pre-training cost for convergence. We evaluated OneDP on 6 challenging simulation tasks as well as 4 self-designed real-world tasks using the Franka robot. The results demonstrate that OneDP not only achieves state-of-the-art success rates but also delivers an order-of-magnitude improvement in inference speed, boosting action prediction frequency from 1.5 Hz to 62 Hz, establishing its potential for dynamic and computationally constrained robotic applications. We share the project page at https://research.nvidia.com/labs/dir/onedp/.

Lu Ling, Chen-Hsuan Lin, Tsung-Yi Lin et al.

Synthesizing interactive 3D scenes from text is essential for gaming, virtual reality, and embodied AI. However, existing methods face several challenges. Learning-based approaches depend on small-scale indoor datasets, limiting the scene diversity and layout complexity. While large language models (LLMs) can leverage diverse text-domain knowledge, they struggle with spatial realism, often producing unnatural object placements that fail to respect common sense. Our key insight is that vision perception can bridge this gap by providing realistic spatial guidance that LLMs lack. To this end, we introduce Scenethesis, a training-free agentic framework that integrates LLM-based scene planning with vision-guided layout refinement. Given a text prompt, Scenethesis first employs an LLM to draft a coarse layout. A vision module then refines it by generating an image guidance and extracting scene structure to capture inter-object relations. Next, an optimization module iteratively enforces accurate pose alignment and physical plausibility, preventing artifacts like object penetration and instability. Finally, a judge module verifies spatial coherence. Comprehensive experiments show that Scenethesis generates diverse, realistic, and physically plausible 3D interactive scenes, making it valuable for virtual content creation, simulation environments, and embodied AI research.

Jiaxiang Tang, Ruijie Lu, Zhaoshuo Li et al.

Recent progress in 3D object generation has greatly improved both the quality and efficiency. However, most existing methods generate a single mesh with all parts fused together, which limits the ability to edit or manipulate individual parts. A key challenge is that different objects may have a varying number of parts. To address this, we propose a new end-to-end framework for part-level 3D object generation. Given a single input image, our method generates high-quality 3D objects with an arbitrary number of complete and semantically meaningful parts. We introduce a dual volume packing strategy that organizes all parts into two complementary volumes, allowing for the creation of complete and interleaved parts that assemble into the final object. Experiments show that our model achieves better quality, diversity, and generalization than previous image-based part-level generation methods.

Zeqi Gu, Yin Cui, Zhaoshuo Li et al.

Designing 3D scenes is traditionally a challenging task that demands both artistic expertise and proficiency with complex software. Recent advances in text-to-3D generation have greatly simplified this process by letting users create scenes based on simple text descriptions. However, as these methods generally require extra training or in-context learning, their performance is often hindered by the limited availability of high-quality 3D data. In contrast, modern text-to-image models learned from web-scale images can generate scenes with diverse, reliable spatial layouts and consistent, visually appealing styles. Our key insight is that instead of learning directly from 3D scenes, we can leverage generated 2D images as an intermediary to guide 3D synthesis. In light of this, we introduce ArtiScene, a training-free automated pipeline for scene design that integrates the flexibility of free-form text-to-image generation with the diversity and reliability of 2D intermediary layouts. First, we generate 2D images from a scene description, then extract the shape and appearance of objects to create 3D models. These models are assembled into the final scene using geometry, position, and pose information derived from the same intermediary image. Being generalizable to a wide range of scenes and styles, ArtiScene outperforms state-of-the-art benchmarks by a large margin in layout and aesthetic quality by quantitative metrics. It also averages a 74.89% winning rate in extensive user studies and 95.07% in GPT-4o evaluation. Project page: https://artiscene-cvpr.github.io/

Zhaoshuo Li, Wei Ye, Dilin Wang et al.

Temporally consistent depth estimation is crucial for online applications such as augmented reality. While stereo depth estimation has received substantial attention as a promising way to generate 3D information, there is relatively little work focused on maintaining temporal stability. Indeed, based on our analysis, current techniques still suffer from poor temporal consistency. Stabilizing depth temporally in dynamic scenes is challenging due to concurrent object and camera motion. In an online setting, this process is further aggravated because only past frames are available. We present a framework named Consistent Online Dynamic Depth (CODD) to produce temporally consistent depth estimates in dynamic scenes in an online setting. CODD augments per-frame stereo networks with novel motion and fusion networks. The motion network accounts for dynamics by predicting a per-pixel SE3 transformation and aligning the observations. The fusion network improves temporal depth consistency by aggregating the current and past estimates. We conduct extensive experiments and demonstrate quantitatively and qualitatively that CODD outperforms competing methods in terms of temporal consistency and performs on par in terms of per-frame accuracy.

Adnan Munawar, Zhaoshuo Li, Punit Kunjam et al.

Surgical simulators not only allow planning and training of complex procedures, but also offer the ability to generate structured data for algorithm development, which may be applied in image-guided computer assisted interventions. While there have been efforts on either developing training platforms for surgeons or data generation engines, these two features, to our knowledge, have not been offered together. We present our developments of a cost-effective and synergistic framework, named Asynchronous Multibody Framework Plus (AMBF+), which generates data for downstream algorithm development simultaneously with users practicing their surgical skills. AMBF+ offers stereoscopic display on a virtual reality (VR) device and haptic feedback for immersive surgical simulation. It can also generate diverse data such as object poses and segmentation maps. AMBF+ is designed with a flexible plugin setup which allows for unobtrusive extension for simulation of different surgical procedures. We show one use case of AMBF+ as a virtual drilling simulator for lateral skull-base surgery, where users can actively modify the patient anatomy using a virtual surgical drill. We further demonstrate how the data generated can be used for validating and training downstream computer vision algorithms

Zhaoshuo Li, Nathan Drenkow, Hao Ding et al.

Scene depth estimation from stereo and monocular imagery is critical for extracting 3D information for downstream tasks such as scene understanding. Recently, learning-based methods for depth estimation have received much attention due to their high performance and flexibility in hardware choice. However, collecting ground truth data for supervised training of these algorithms is costly or outright impossible. This circumstance suggests a need for alternative learning approaches that do not require corresponding depth measurements. Indeed, self-supervised learning of depth estimation provides an increasingly popular alternative. It is based on the idea that observed frames can be synthesized from neighboring frames if accurate depth of the scene is known - or in this case, estimated. We show empirically that - contrary to common belief - improvements in image synthesis do not necessitate improvement in depth estimation. Rather, optimizing for image synthesis can result in diverging performance with respect to the main prediction objective - depth. We attribute this diverging phenomenon to aleatoric uncertainties, which originate from data. Based on our experiments on four datasets (spanning street, indoor, and medical) and five architectures (monocular and stereo), we conclude that this diverging phenomenon is independent of the dataset domain and not mitigated by commonly used regularization techniques. To underscore the importance of this finding, we include a survey of methods which use image synthesis, totaling 127 papers over the last six years. This observed divergence has not been previously reported or studied in depth, suggesting room for future improvement of self-supervised approaches which might be impacted the finding.

Yonghao Long, Zhaoshuo Li, Chi Hang Yee et al.

Reconstructing the scene of robotic surgery from the stereo endoscopic video is an important and promising topic in surgical data science, which potentially supports many applications such as surgical visual perception, robotic surgery education and intra-operative context awareness. However, current methods are mostly restricted to reconstructing static anatomy assuming no tissue deformation, tool occlusion and de-occlusion, and camera movement. However, these assumptions are not always satisfied in minimal invasive robotic surgeries. In this work, we present an efficient reconstruction pipeline for highly dynamic surgical scenes that runs at 28 fps. Specifically, we design a transformer-based stereoscopic depth perception for efficient depth estimation and a light-weight tool segmentor to handle tool occlusion. After that, a dynamic reconstruction algorithm which can estimate the tissue deformation and camera movement, and aggregate the information over time is proposed for surgical scene reconstruction. We evaluate the proposed pipeline on two datasets, the public Hamlyn Centre Endoscopic Video Dataset and our in-house DaVinci robotic surgery dataset. The results demonstrate that our method can recover the scene obstructed by the surgical tool and handle the movement of camera in realistic surgical scenarios effectively at real-time speed.

Anna Zapaishchykova, David Dreizin, Zhaoshuo Li et al.

Pelvic ring disruptions result from blunt injury mechanisms and are often found in patients with multi-system trauma. To grade pelvic fracture severity in trauma victims based on whole-body CT, the Tile AO/OTA classification is frequently used. Due to the high volume of whole-body trauma CTs generated in busy trauma centers, an automated approach to Tile classification would provide substantial value, e.,g., to prioritize the reading queue of the attending trauma radiologist. In such scenario, an automated method should perform grading based on a transparent process and based on interpretable features to enable interaction with human readers and lower their workload by offering insights from a first automated read of the scan. This paper introduces an automated yet interpretable pelvic trauma decision support system to assist radiologists in fracture detection and Tile grade classification. The method operates similarly to human interpretation of CT scans and first detects distinct pelvic fractures on CT with high specificity using a Faster-RCNN model that are then interpreted using a structural causal model based on clinical best practices to infer an initial Tile grade. The Bayesian causal model and finally, the object detector are then queried for likely co-occurring fractures that may have been rejected initially due to the highly specific operating point of the detector, resulting in an updated list of detected fractures and corresponding final Tile grade. Our method is transparent in that it provides finding location and type using the object detector, as well as information on important counterfactuals that would invalidate the system's recommendation and achieves an AUC of 83.3%/85.1% for translational/rotational instability. Despite being designed for human-machine teaming, our approach does not compromise on performance compared to previous black-box approaches.

Zhaoshuo Li, Xingtong Liu, Nathan Drenkow et al.

Stereo depth estimation relies on optimal correspondence matching between pixels on epipolar lines in the left and right images to infer depth. In this work, we revisit the problem from a sequence-to-sequence correspondence perspective to replace cost volume construction with dense pixel matching using position information and attention. This approach, named STereo TRansformer (STTR), has several advantages: It 1) relaxes the limitation of a fixed disparity range, 2) identifies occluded regions and provides confidence estimates, and 3) imposes uniqueness constraints during the matching process. We report promising results on both synthetic and real-world datasets and demonstrate that STTR generalizes across different domains, even without fine-tuning.

Zhaoshuo Li, Amirreza Shaban, Jean-Gabriel Simard et al.

Purpose: We describe a 3D multi-view perception system for the da Vinci surgical system to enable Operating room (OR) scene understanding and context awareness. Methods: Our proposed system is comprised of four Time-of-Flight (ToF) cameras rigidly attached to strategic locations on the daVinci Xi patient side cart (PSC). The cameras are registered to the robot's kinematic chain by performing a one-time calibration routine and therefore, information from all cameras can be fused and represented in one common coordinate frame. Based on this architecture, a multi-view 3D scene semantic segmentation algorithm is created to enable recognition of common and salient objects/equipment and surgical activities in a da Vinci OR. Our proposed 3D semantic segmentation method has been trained and validated on a novel densely annotated dataset that has been captured from clinical scenarios. Results: The results show that our proposed architecture has acceptable registration error ($3.3\%\pm1.4\%$ of object-camera distance) and can robustly improve scene segmentation performance (mean Intersection Over Union - mIOU) for less frequently appearing classes ($\ge 0.013$) compared to a single-view method. Conclusion: We present the first dynamic multi-view perception system with a novel segmentation architecture, which can be used as a building block technology for applications such as surgical workflow analysis, automation of surgical sub-tasks and advanced guidance systems.

Zhaoshuo Li, Mahya Shahbazi, Niravkumar Patel et al.

High-resolution real-time intraocular imaging of retina at the cellular level is very challenging due to the vulnerable and confined space within the eyeball as well as the limited availability of appropriate modalities. A probe-based confocal laser endomicroscopy (pCLE) system, can be a potential imaging modality for improved diagnosis. The ability to visualize the retina at the cellular level could provide information that may predict surgical outcomes. The adoption of intraocular pCLE scanning is currently limited due to the narrow field of view and the micron-scale range of focus. In the absence of motion compensation, physiological tremors of the surgeons' hand and patient movements also contribute to the deterioration of the image quality. Therefore, an image-based hybrid control strategy is proposed to mitigate the above challenges. The proposed hybrid control strategy enables a shared control of the pCLE probe between surgeons and robots to scan the retina precisely, with the absence of hand tremors and with the advantages of an image-based auto-focus algorithm that optimizes the quality of pCLE images. The hybrid control strategy is deployed on two frameworks - cooperative and teleoperated. Better image quality, smoother motion, and reduced workload are all achieved in a statistically significant manner with the hybrid control frameworks.