Activity
2K followers
-
Gavriel State reposted thisGavriel State reposted this🚀 𝗘𝘅𝗰𝗶𝘁𝗲𝗱 𝘁𝗼 𝗮𝗻𝗻𝗼𝘂𝗻𝗰𝗲 𝘁𝗵𝗮𝘁 𝘁𝗵𝗲 𝗰𝗼𝗱𝗲 𝗮𝗻𝗱 𝗺𝗼𝗱𝗲𝗹𝘀 𝗳𝗼𝗿 𝗼𝘂𝗿 𝗜𝗖𝗟𝗥 𝟮𝟬𝟮𝟲 𝗽𝗮𝗽𝗲𝗿, 𝗩𝗼𝗠𝗣, 𝗮𝗿𝗲 𝗻𝗼𝘄 𝗼𝗽𝗲𝗻-𝘀𝗼𝘂𝗿𝗰𝗲! Physical simulation has always had a bottleneck: the laborious, hand-crafted assignment of spatially varying mechanical properties. To solve this, we’re introducing 𝗩𝗼𝗠𝗣 (𝗣𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗻𝗴 𝗩𝗼𝗹𝘂𝗺𝗲𝘁𝗿𝗶𝗰 𝗠𝗲𝗰𝗵𝗮𝗻𝗶𝗰𝗮𝗹 𝗣𝗿𝗼𝗽𝗲𝗿𝘁𝘆 𝗙𝗶𝗲𝗹𝗱𝘀)—the first feed-forward model to predict fine-grained mechanical properties throughout the volume of 3D objects. If you are working in 𝗥𝗼𝗯𝗼𝘁𝗶𝗰𝘀, 𝗗𝗶𝗴𝗶𝘁𝗮𝗹 𝗧𝘄𝗶𝗻𝘀, 𝗼𝗿 𝗥𝗲𝗮𝗹𝟮𝗦𝗶𝗺, VoMP can automatically convert your 3D representations into simulation-ready assets. 𝗞𝗲𝘆 𝗛𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁𝘀: 🧠 𝗨𝗻𝗶𝘃𝗲𝗿𝘀𝗮𝗹 𝟯𝗗 𝗖𝗼𝗺𝗽𝗮𝘁𝗶𝗯𝗶𝗹𝗶𝘁𝘆: Works seamlessly with any renderable/voxelizable 3D representation, including meshes, Gaussian Splats, and SDFs. ⚙️ 𝗖𝗼𝗺𝗽𝗿𝗲𝗵𝗲𝗻𝘀𝗶𝘃𝗲 𝗣𝗵𝘆𝘀𝗶𝗰𝘀: Predicts per-voxel Young's modulus (E), Poisson's ratio (ν), and density (ρ) for realistic deformable simulations. 📐 𝗣𝗵𝘆𝘀𝗶𝗰𝗮𝗹𝗹𝘆 𝗣𝗹𝗮𝘂𝘀𝗶𝗯𝗹𝗲: Uses a Geometry Transformer to predict material latents that reside on a trained manifold of real-world materials, guaranteeing validity. 📊 𝗡𝗼𝘃𝗲𝗹 𝗣𝗶𝗽𝗲𝗹𝗶𝗻𝗲: Trained using a new annotation pipeline that fuses segmented 3D datasets, material databases, and VLMs. We’re thrilled to see how the community uses this to push the boundaries of physics-based simulation! 🌐 𝗣𝗿𝗼𝗷𝗲𝗰𝘁 𝗣𝗮𝗴𝗲: https://lnkd.in/g5bcZ4wd Kudos Rishit Dagli, Donglai Xiang, Vismay M., Charles Loop, Clement Fuji Tsang, Anka He Chen, Anita Hu, Gavriel State, David I.W. Levin!
-
Gavriel State reposted thisWe're proud to release our work on "PPISP: Physically-Plausible Compensation and Control of Photometric Variations in Radiance Field Reconstruction" today! Michael Rubloff already tried it out, see his post below! Many thanks to the team: Nicolas Moënne-Loccoz, Gavriel State, and Zan Gojcic! Project website: https://lnkd.in/d3fC5ax5 Code: https://lnkd.in/dxM8zNr8 Paper: https://lnkd.in/d9isV87mGavriel State reposted thisToday's a big day for radiance fields like gaussian splatting! If you've been struggling with floaters or inconsistent color, NVIDIA AI just released Physically-Plausible Image Signal Processing (PPISP) for Radiance Field Reconstruction. PPISP explicitly models the camera’s image pipeline inside radiance field reconstructions, then learning auto exposure and auto white balance behavior for novel views. It's also a generalized radiance field method, meaning it will work with NeRFs, gaussian splatting, 3DGUT, voxels, and more! The code is also released with an Apache 2.0 license and will be coming to repositories like gsplat and 3DGRUT soon! For people who are more familiar, this benchmarks directly against Bilagrid. Article: https://lnkd.in/e-C_Z7aY Code: https://lnkd.in/eVMkMrRD Zan Gojcic Isaac Deutsch
-
Gavriel State shared thisGreat work from Rishit showing how we can apply physics properties to scanned objects! This will be a great help in robot simulation to help reduce sim2real gaps!Gavriel State shared this📣 want to produce realistic dynamic 3d worlds (with >100 splats) our new project, VoMP, is the first feed forward approach to convert input surface geometry to volumetric sim-ready assets by assigning real world physics materials using physical simulation for producing dynamic 3d scenes with rich realistic interaction relies on spatially-varying physically-based mechanical properties throughout the volume of the object. these are typically laboriously hand-crafted for every object with much trial-error with VoMP we can now build realistic dynamic 3D interactive worlds powered by VoMP properties: - make a 3d gaussian splat or mesh or other representation environment interactive and run a robot through it, or - simulate dynamic 3D worlds with hundreds of deformable objects with collisions 🌐 Project: https://lnkd.in/eMj8m2DK 📜 Paper: https://lnkd.in/e7cTmv8s
-
Gavriel State reposted thisGavriel State reposted thisBuilding realistic 3D environments for robotics simulation can be a labor-intensive process. Now, with NVIDIA Omniverse NuRec, you can complete the entire process using just a smartphone. Learn how ➡️ https://bit.ly/49iyOXu
-
Gavriel State reposted thisGavriel State reposted thisExtremely proud of all that have contributed to the Isaac Lab project. Read all about our approach to next generation simulation-based robot learning in this comprehensive whitepaper! https://lnkd.in/gxCy4mkB
-
Gavriel State shared thisI’m looking forward to chatting with anyone going to CoRL about Isaac Lab and Newton! The meetings we had at CoRL last year were key to some of the decisions we’ve made since then, including starting the Newton project. Sign up below!Gavriel State shared thisHeaded to #CoRL2025 in Seoul? 👋 Meet the NVIDIA team and explore how #NVIDIAIsaac can accelerate your next robotics project. Book a time with us ➡️ https://nvda.ws/3VXs7lM
-
Gavriel State reposted thisGavriel State reposted thisUsing Isaac Sim to hammer a nail to the wall to hang a photo. Trying to hang a photo to the wall with LeRobot and Isaac Sim. This is teleoperation, using https://lnkd.in/giQDZZ5m Smart home robot training is becoming easier with these assets!
-
Gavriel State reposted thisGavriel State reposted thisWe’re standing at the intersection of human ingenuity and machine intelligence. For years, developers have been pushing the boundaries of what’s possible in AI and robotics but limited by the computing power available on the edge. Today, that changes. Today, we introduce the NVIDIA Jetson Thor. Think different. Thor isn’t just another chip; it’s a revolution in edge AI computing, designed from the ground up for the physical world - a real-time reasoning machine. Built on the groundbreaking Blackwell architecture, it packs 2,560 CUDA cores and 96 fifth-generation Tensor Cores, delivering up to 2,070 TFLOPS of FP4 AI performance—that’s 7.5 times the AI muscle of its predecessor, the AGX Orin, with 3.5 times the efficiency. It comes with With a 14-core Arm Neoverse-V3AE CPU, 128 GB of lightning-fast LPDDR5X memory to run the largest reasoning models and complex control functions. But here’s what excites me most: Thor makes development magical. We’ve integrated it seamlessly with NVIDIA’s AI ecosystem—CUDA, TensorRT, and the JetPack SDK—so you can prototype, train, and deploy AI models faster than ever. Cosmos Reason, DeepSeek, Llama, Gemini, Qwen, Isaac GR00T and many other open source foundation models run out of the box. No more wrestling with incompatible tools or inefficient workflows. Thor empowers you to create experiences that feel alive, intuitive, and transformative. I am very proud of my NVIDIA Robotics team that has worked relentlessly to bring this amazing product to market. We are eagerly waiting to see what you all will build next. With Jetson Thor, it’s not just possible—it’s inevitable!
-
Gavriel State reposted thisGavriel State reposted thisWe are excited to be among the very first groups selected by NVIDIA Robotics to test the new NVIDIA #Thor. We have managed to run a #VisionLanguageModel (Qwen 2.5 VL) for semantic understanding of the environment, along with a monocular depth model (#DepthAnything v2), for safe autonomous navigation, all onboard, no cloud, no internet connection required! The video shows a simple result obtained in just two weeks of work. Kudos to Leonard Bauersfeld Jiaxu Xing Ismail Geles Yannick Armati for making this possible! #ComputerVision #Robotics University of Zurich Faculty of Science University of Zurich UZH Department of Informatics European Research Council (ERC) Switzerland Innovation Park Zurich
-
Gavriel State liked thisGavriel State liked this🚀 𝗘𝘅𝗰𝗶𝘁𝗲𝗱 𝘁𝗼 𝗮𝗻𝗻𝗼𝘂𝗻𝗰𝗲 𝘁𝗵𝗮𝘁 𝘁𝗵𝗲 𝗰𝗼𝗱𝗲 𝗮𝗻𝗱 𝗺𝗼𝗱𝗲𝗹𝘀 𝗳𝗼𝗿 𝗼𝘂𝗿 𝗜𝗖𝗟𝗥 𝟮𝟬𝟮𝟲 𝗽𝗮𝗽𝗲𝗿, 𝗩𝗼𝗠𝗣, 𝗮𝗿𝗲 𝗻𝗼𝘄 𝗼𝗽𝗲𝗻-𝘀𝗼𝘂𝗿𝗰𝗲! Physical simulation has always had a bottleneck: the laborious, hand-crafted assignment of spatially varying mechanical properties. To solve this, we’re introducing 𝗩𝗼𝗠𝗣 (𝗣𝗿𝗲𝗱𝗶𝗰𝘁𝗶𝗻𝗴 𝗩𝗼𝗹𝘂𝗺𝗲𝘁𝗿𝗶𝗰 𝗠𝗲𝗰𝗵𝗮𝗻𝗶𝗰𝗮𝗹 𝗣𝗿𝗼𝗽𝗲𝗿𝘁𝘆 𝗙𝗶𝗲𝗹𝗱𝘀)—the first feed-forward model to predict fine-grained mechanical properties throughout the volume of 3D objects. If you are working in 𝗥𝗼𝗯𝗼𝘁𝗶𝗰𝘀, 𝗗𝗶𝗴𝗶𝘁𝗮𝗹 𝗧𝘄𝗶𝗻𝘀, 𝗼𝗿 𝗥𝗲𝗮𝗹𝟮𝗦𝗶𝗺, VoMP can automatically convert your 3D representations into simulation-ready assets. 𝗞𝗲𝘆 𝗛𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁𝘀: 🧠 𝗨𝗻𝗶𝘃𝗲𝗿𝘀𝗮𝗹 𝟯𝗗 𝗖𝗼𝗺𝗽𝗮𝘁𝗶𝗯𝗶𝗹𝗶𝘁𝘆: Works seamlessly with any renderable/voxelizable 3D representation, including meshes, Gaussian Splats, and SDFs. ⚙️ 𝗖𝗼𝗺𝗽𝗿𝗲𝗵𝗲𝗻𝘀𝗶𝘃𝗲 𝗣𝗵𝘆𝘀𝗶𝗰𝘀: Predicts per-voxel Young's modulus (E), Poisson's ratio (ν), and density (ρ) for realistic deformable simulations. 📐 𝗣𝗵𝘆𝘀𝗶𝗰𝗮𝗹𝗹𝘆 𝗣𝗹𝗮𝘂𝘀𝗶𝗯𝗹𝗲: Uses a Geometry Transformer to predict material latents that reside on a trained manifold of real-world materials, guaranteeing validity. 📊 𝗡𝗼𝘃𝗲𝗹 𝗣𝗶𝗽𝗲𝗹𝗶𝗻𝗲: Trained using a new annotation pipeline that fuses segmented 3D datasets, material databases, and VLMs. We’re thrilled to see how the community uses this to push the boundaries of physics-based simulation! 🌐 𝗣𝗿𝗼𝗷𝗲𝗰𝘁 𝗣𝗮𝗴𝗲: https://lnkd.in/g5bcZ4wd Kudos Rishit Dagli, Donglai Xiang, Vismay M., Charles Loop, Clement Fuji Tsang, Anka He Chen, Anita Hu, Gavriel State, David I.W. Levin!
-
Gavriel State liked thisGavriel State liked thisLast week in Bangkok, I had the privilege of joining fellow leaders at the YPO Global Leadership Meetings, an experience that reinforced both the responsibility and opportunity we have to shape the future of our global CEO community. As a member of the YPO Global Learning Committee, our work goes far beyond programming. Over several days, we engaged in rigorous dialogue and alignment to forge a multi-year strategy that will elevate how learning is designed, delivered, and experienced across YPO. This includes rethinking how we create transformative, high-impact learning journeys that remain relevant for 35,000+ members navigating an increasingly complex world. These meetings are where the future of YPO is actively built, where ideas are tested, challenged, and ultimately translated into initiatives that will influence how leaders grow, connect, and lead across every region. Equally important, we remain students ourselves. A standout session with Harvard University Professor Boris Groysberg offered powerful insights on leadership, talent, and what truly differentiates high-performing organizations, reminding us that great leadership starts with lifelong learning. I left Bangkok both energized and deeply committed to my work at YPO. It is a privilege to contribute to an organization that is not only global in reach, but intentional in its pursuit of meaningful, lifelong learning at the highest level. #YPO #GlobalLeadership #Leadership #LifelongLearning #ExecutiveLeadership #HighPerformance #YPOGlobal #LeadershipDevelopment #HarvardBusinessSchool #CEOCommunity #Impact
-
Gavriel State liked thisGavriel State liked thisI recently decided to pause my PhD and focus on building my career at NVIDIA. I’m excited to share that I started full-time in February as a Robotics Simulation Software Engineer on the Isaac Sim team. 💚 This announcement comes just after my first NVIDIA GTC, where I had the opportunity to: 🔹 co-instruct a hands-on training lab, “Configure and Tune Robot Assets with #OpenUSD and #PhysX”, with Ji Yuan Feng 🔹 serve as an expert in a Connect with Experts session on OpenUSD and Physical AI development 🔹 help with the Simulate Complex Robot-to-Object Interactions booth demo (S/O Ayush Ghosh) Feeling grateful to have been trusted to contribute across so many parts of #GTC26 this early. It was a rewarding and fun way to get involved and connect with the people building in this space! I’m especially grateful to the Isaac Sim team, colleagues across NVIDIA I've had the chance to work or connect with, mentors, and manager for the trust and support. Special shoutouts to: 🔹 Renato Gasoto and my manager Hammad Mazhar for being great coaches and mentors, and for the opportunity to join the team 🔹 My grad school advisor Dan Negrut for the unwavering support and sound advice over the years 🔹 Our awesome learning PIC Shane Reetz for his tireless work making sure the robotics training labs would run smoothly Looking forward to what’s next 💚
-
Gavriel State liked thisGavriel State liked thisNewton 1.0 is live. GA release highlights: 🧩 Stable, unified API across modeling, solving, control, and sensing ⚡ MuJoCo Warp solver: up to 252× (locomotion) and 475× (manipulation) speedups for MJX 🤖 Kamino solver (beta): complex mechanisms including linkages and loop closures 🧵 Deformable solvers: VBD + MPM, explicit two-way coupling with MuJoCo Warp 🎯 SDF collisions + hydroelastic contact for high-fidelity interactions 🔗 Isaac Lab and Sim early access + OpenUSD integration for end-to-end robot learning 👁️ Tiled camera sensor fully written in Warp for high-throughput vision-based RL on DGX platform Post-1.0 release and feature roadmap: Monthly releases rolling out multiphysics features (automatic coupling, impulse exchange API, richer behaviors), standardized USD schemas, a multiphysics asset library, and advanced throughput-optimized solvers, while evaluating low-latency paths, faster CPU execution, deterministic simulation, and broader differentiability. 👉 Full release + roadmap: https://lnkd.in/gntj-GqR Built in the open, shaped with the community. Looking forward to what you build with Newton. Example below shows Ethernet cable manipulation and RJ45 snap-fit insertion in the context of GB300 GPU assembly. The cable and rigid bodies are simulated with Newton SDF collisions/contacts and VBD solver, visualized in Newton viewer.
-
Gavriel State liked thisGavriel State liked thisIf you’re at GDC this week, come check out a live demo of the upcoming vk_gaussian_splating (https://lnkd.in/e4RtSDRR) release—presented ahead of publication (coming soon). What’s new: 1. Support for very large 3DGS scenes across all pipelines: raster, ray tracing, and hybrid 2. Real-time ray tracing with lighting and shadows Come see it live—I won’t be on-site, but my teammates will be at the NVIDIA booth and happy to welcome you, show the demo, and answer questions. City and Greenhouse models courtesy Andrii Shramko https://lnkd.in/eFX_SXFz
-
Gavriel State liked thisGavriel State liked thisWhen Specs Save the Day 🔍 Over the past week, I scratched out ~4 hours of spare time across two AI-assisted sessions to build a parser bringing Niantic's SPZ files (compressed 3D Gaussian Splats) into systems literate with OpenUSD v26.03's particle fields schema. Several bugs in the initial attempt-- every one traced to spec details that were initially skimmed over: ❌ Wrong field order ❌ Values stored post-transform, not raw ❌ "Colors" were actually SH coefficients ❌ Packed quaternions use "smallest three" encoding Hits different when you're AOUSD Core Specification Working Group Chair. Every ambiguity I leave in a spec is a bug someone else ships. The silver lining? Cursor/Claude + Codex took us from concept → broken parser → working converter → documented postmortem in those 4 hours. Fast feedback makes spec gaps visible faster, so we can fix them for everyone. Back to reviewing schema docs with fresh eyes. 👀 #OpenUSD #Specifications #AOUSD #LessonsLearned #3DGS #OpenStandards
-
Gavriel State liked thisWe're proud to release our work on "PPISP: Physically-Plausible Compensation and Control of Photometric Variations in Radiance Field Reconstruction" today! Michael Rubloff already tried it out, see his post below! Many thanks to the team: Nicolas Moënne-Loccoz, Gavriel State, and Zan Gojcic! Project website: https://lnkd.in/d3fC5ax5 Code: https://lnkd.in/dxM8zNr8 Paper: https://lnkd.in/d9isV87mGavriel State liked thisToday's a big day for radiance fields like gaussian splatting! If you've been struggling with floaters or inconsistent color, NVIDIA AI just released Physically-Plausible Image Signal Processing (PPISP) for Radiance Field Reconstruction. PPISP explicitly models the camera’s image pipeline inside radiance field reconstructions, then learning auto exposure and auto white balance behavior for novel views. It's also a generalized radiance field method, meaning it will work with NeRFs, gaussian splatting, 3DGUT, voxels, and more! The code is also released with an Apache 2.0 license and will be coming to repositories like gsplat and 3DGRUT soon! For people who are more familiar, this benchmarks directly against Bilagrid. Article: https://lnkd.in/e-C_Z7aY Code: https://lnkd.in/eVMkMrRD Zan Gojcic Isaac Deutsch
Experience & Education
-
NVIDIA
View Gavriel’s full experience
By clicking Continue to join or sign in, you agree to LinkedIn’s User Agreement, Privacy Policy, and Cookie Policy.
Patents
-
General purpose software parallel task engine
Issued US 8284206
For TransGaming's SwiftShader software 3D renderer, we developed a technique for doing dynamic code generation for parallel execution across multiple cores.
Other inventors -
Recommendations received
1 person has recommended Gavriel
Join now to viewOther similar profiles
Explore more posts
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore More