arXiv subjects

Aryaman Gupta

Publications and source records attributed to Aryaman Gupta.

13 recordsLinked to original sources

arXiv2026

Cosmos 3: Omnimodal World Models for Physical AI

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 License at https://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.

cs.CV↗

arXiv2025

World Simulation with Video Foundation Models for Physical AI

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.

cs.CV↗

arXiv2025

Unsupervised Discovery of Failure Taxonomies from Deployment Logs

As robotic systems become increasingly integrated into real-world environments, ranging from autonomous vehicles to household assistants, they inevitably encounter diverse and unstructured scenarios that lead to failures. While such failures pose safety and reliability challenges, they also provide rich perceptual data for improving system robustness. However, manually analyzing large-scale failure datasets is impractical and does not scale. In this work, we introduce the problem of unsupervised discovery of failure taxonomies from large volumes of raw failure logs, aiming to obtain semantically coherent and actionable failure modes directly from perceptual trajectories. Our approach first infers structured failure explanations from multimodal inputs using vision language reasoning, then clusters them in the resulting semantic reasoning space, discovering recurring failure modes rather than isolated episode-level descriptions. We evaluate our method across robotic manipulation, indoor navigation, and autonomous driving domains, demonstrating that the discovered taxonomies are consistent, interpretable, and useful in practice. In particular, we show that structured failure taxonomies guide targeted data collection for offline policy refinement and enhance runtime failure monitoring systems. Website: https://mllm-failure-clustering.github.io/

cs.RO↗

arXiv2024

Fourier Domain Adaptation for Traffic Light Detection in Adverse Weather

Traffic light detection under adverse weather conditions remains largely unexplored in ADAS systems, with existing approaches relying on complex deep learning methods that introduce significant computational overheads during training and deployment. This paper proposes Fourier Domain Adaptation (FDA), which requires only training data modifications without architectural changes, enabling effective adaptation to rainy and foggy conditions. FDA minimizes the domain gap between source and target domains, creating a dataset for reliable performance under adverse weather. The source domain merged LISA and S2TLD datasets, processed to address class imbalance. Established methods simulated rainy and foggy scenarios to form the target domain. Semi-Supervised Learning (SSL) techniques were explored to leverage data more effectively, addressing the shortage of comprehensive datasets and poor performance of state-of-the-art models under hostile weather. Experimental results show FDA-augmented models outperform baseline models across mAP50, mAP50-95, Precision, and Recall metrics. YOLOv8 achieved a 12.25% average increase across all metrics. Average improvements of 7.69% in Precision, 19.91% in Recall, 15.85% in mAP50, and 23.81% in mAP50-95 were observed across all models, demonstrating FDA's effectiveness in mitigating adverse weather impact. These improvements enable real-world applications requiring reliable performance in challenging environmental conditions.

cs.CV↗

arXiv2024

Enhancing Safety and Robustness of Vision-Based Controllers via Reachability Analysis

Autonomous systems, such as self-driving cars and drones, have made significant strides in recent years by leveraging visual inputs and machine learning for decision-making and control. Despite their impressive performance, these vision-based controllers can make erroneous predictions when faced with novel or out-of-distribution inputs. Such errors can cascade into catastrophic system failures and compromise system safety. In this work, we compute Neural Reachable Tubes, which act as parameterized approximations of Backward Reachable Tubes to stress-test the vision-based controllers and mine their failure modes. The identified failures are then used to enhance the system safety through both offline and online methods. The online approach involves training a classifier as a run-time failure monitor to detect closed-loop, system-level failures, subsequently triggering a fallback controller that robustly handles these detected failures to preserve system safety. For the offline approach, we improve the original controller via incremental training using a carefully augmented failure dataset, resulting in a more robust controller that is resistant to the known failure modes. In either approach, the system is safeguarded against shortcomings that transcend the vision-based controller and pertain to the closed-loop safety of the overall system. We validate the proposed approaches on an autonomous aircraft taxiing task that involves using a vision-based controller to guide the aircraft towards the centerline of the runway. Our results show the efficacy of the proposed algorithms in identifying and handling system-level failures, outperforming methods that rely on controller prediction error or uncertainty quantification for identifying system failures.

cs.RO↗

arXiv2024

Deep Reinforcement Learning for Sim-to-Real Policy Transfer of VTOL-UAVs Offshore Docking Operations

This paper proposes a novel Reinforcement Learning (RL) approach for sim-to-real policy transfer of Vertical Take-Off and Landing Unmanned Aerial Vehicle (VTOL-UAV). The proposed approach is designed for VTOL-UAV landing on offshore docking stations in maritime operations. VTOL-UAVs in maritime operations encounter limitations in their operational range, primarily stemming from constraints imposed by their battery capacity. The concept of autonomous landing on a charging platform presents an intriguing prospect for mitigating these limitations by facilitating battery charging and data transfer. However, current Deep Reinforcement Learning (DRL) methods exhibit drawbacks, including lengthy training times, and modest success rates. In this paper, we tackle these concerns comprehensively by decomposing the landing procedure into a sequence of more manageable but analogous tasks in terms of an approach phase and a landing phase. The proposed architecture utilizes a model-based control scheme for the approach phase, where the VTOL-UAV is approaching the offshore docking station. In the Landing phase, DRL agents were trained offline to learn the optimal policy to dock on the offshore station. The Joint North Sea Wave Project (JONSWAP) spectrum model has been employed to create a wave model for each episode, enhancing policy generalization for sim2real transfer. A set of DRL algorithms have been tested through numerical simulations including value-based agents and policy-based agents such as Deep \textit{Q} Networks (DQN) and Proximal Policy Optimization (PPO) respectively. The numerical experiments show that the PPO agent can learn complicated and efficient policies to land in uncertain environments, which in turn enhances the likelihood of successful sim-to-real transfer.

cs.RO↗

arXiv2023

Detecting and Mitigating System-Level Anomalies of Vision-Based Controllers

Autonomous systems, such as self-driving cars and drones, have made significant strides in recent years by leveraging visual inputs and machine learning for decision-making and control. Despite their impressive performance, these vision-based controllers can make erroneous predictions when faced with novel or out-of-distribution inputs. Such errors can cascade to catastrophic system failures and compromise system safety. In this work, we introduce a run-time anomaly monitor to detect and mitigate such closed-loop, system-level failures. Specifically, we leverage a reachability-based framework to stress-test the vision-based controller offline and mine its system-level failures. This data is then used to train a classifier that is leveraged online to flag inputs that might cause system breakdowns. The anomaly detector highlights issues that transcend individual modules and pertain to the safety of the overall system. We also design a fallback controller that robustly handles these detected anomalies to preserve system safety. We validate the proposed approach on an autonomous aircraft taxiing system that uses a vision-based controller for taxiing. Our results show the efficacy of the proposed approach in identifying and handling system-level anomalies, outperforming methods such as prediction error-based detection, and ensembling, thereby enhancing the overall safety and robustness of autonomous systems.

cs.RO↗

arXiv2023

Efficient Raycasting of Volumetric Depth Images for Remote Visualization of Large Volumes at High Frame Rates

We present an efficient raycasting algorithm for rendering Volumetric Depth Images (VDIs), and we show how it can be used in a remote visualization setting with VDIs generated and streamed from a remote server. VDIs are compact view-dependent volume representations that enable interactive visualization of large volumes at high frame rates by decoupling viewpoint changes from expensive rendering calculations. However, current rendering approaches for VDIs struggle with achieving interactive frame rates at high image resolutions. Here, we exploit the properties of perspective projection to simplify intersections of rays with the view-dependent frustums in a VDI and leverage spatial smoothness in the volume data to minimize memory accesses. Benchmarks show that responsive frame rates can be achieved close to the viewpoint of generation for HD display resolutions, providing high-fidelity approximate renderings of Gigabyte-sized volumes. We also propose a method to subsample the VDI for preview rendering, maintaining high frame rates even for large viewpoint deviations. We provide our implementation as an extension of an established open-source visualization library.

cs.GR↗

arXiv2022

Parallel Compositing of Volumetric Depth Images for Interactive Visualization of Distributed Volumes at High Frame Rates

We present a parallel compositing algorithm for Volumetric Depth Images (VDIs) of large three-dimensional volume data. Large distributed volume data are routinely produced in both numerical simulations and experiments, yet it remains challenging to visualize them at smooth, interactive frame rates. VDIs are view-dependent piecewise constant representations of volume data that offer a potential solution. They are more compact and less expensive to render than the original data. So far, however, there is no method for generating VDIs from distributed data. We propose an algorithm that enables this by sort-last parallel generation and compositing of VDIs with automatically chosen content-adaptive parameters. The resulting composited VDI can then be streamed for remote display, providing responsive visualization of large, distributed volume data.

cs.GR↗

arXiv2021

Inscribed triangles of Jordan curves in $\mathbb{R}^{n}$

Nielsen's theorem states that any triangle can be inscribed in a planar Jordan curve. We prove a generalisation of this theorem, extending to any Jordan curve $J$ embedded in $\mathbb{R}^{n}$, for a restricted set of triangles. We then conclude by investigating a condition under which a given point of $J$ inscribes an equilateral triangle in particular.

math.MG↗

arXiv2020

scenery: Flexible Virtual Reality Visualization on the Java VM

Life science today involves computational analysis of a large amount and variety of data, such as volumetric data acquired by state-of-the-art microscopes, or mesh data from analysis of such data or simulations. Visualization is often the first step in making sense of data, and a crucial part of building and debugging analysis pipelines. It is therefore important that visualizations can be quickly prototyped, as well as developed or embedded into full applications. In order to better judge spatiotemporal relationships, immersive hardware, such as Virtual or Augmented Reality (VR/AR) headsets and associated controllers are becoming invaluable tools. In this work we introduce scenery, a flexible VR/AR visualization framework for the Java VM that can handle mesh and large volumetric data, containing multiple views, timepoints, and color channels. scenery is free and open-source software, works on all major platforms, and uses the Vulkan or OpenGL rendering APIs. We introduce scenery's main features and example applications, such as its use in VR for microscopy, in the biomedical image analysis software Fiji, or for visualizing agent-based simulations.

cs.GR↗

arXiv2019

A Proposed Framework for Interactive Virtual Reality In Situ Visualization of Parallel Numerical Simulations

As computer simulations progress to increasingly complex, non-linear, and three-dimensional systems and phenomena, intuitive and immediate visualization of their results is becoming crucial. While Virtual Reality (VR) and Natural User Interfaces (NUIs) have been shown to improve understanding of complex 3D data, their application to live in situ visualization and computational steering is hampered by performance requirements. Here, we present the design of a software framework for interactive VR in situ visualization of parallel numerical simulations, as well as a working prototype implementation. Our design is targeted towards meeting the performance requirements for VR, and our work is packaged in a framework that allows for easy instrumentation of simulations. Our preliminary results inform about the technical feasibility of the architecture, as well as the challenges that remain.

cs.DC↗

arXiv2018

Nose, eyes and ears: Head pose estimation by locating facial keypoints

Monocular head pose estimation requires learning a model that computes the intrinsic Euler angles for pose (yaw, pitch, roll) from an input image of human face. Annotating ground truth head pose angles for images in the wild is difficult and requires ad-hoc fitting procedures (which provides only coarse and approximate annotations). This highlights the need for approaches which can train on data captured in controlled environment and generalize on the images in the wild (with varying appearance and illumination of the face). Most present day deep learning approaches which learn a regression function directly on the input images fail to do so. To this end, we propose to use a higher level representation to regress the head pose while using deep learning architectures. More specifically, we use the uncertainty maps in the form of 2D soft localization heatmap images over five facial keypoints, namely left ear, right ear, left eye, right eye and nose, and pass them through an convolutional neural network to regress the head-pose. We show head pose estimation results on two challenging benchmarks BIWI and AFLW and our approach surpasses the state of the art on both the datasets.

cs.CV↗