We'll present Déjà View as Oral at #NeurIPS 🎉 TL;DR: we beat VGGT-Ω on average with a 10x smaller, looped architecture. Check it out 👇

Claude-Pop - I'm Upping My P(Doom)

Introducing OM-1, our first robot foundation model, zero-shot generalizing to any robot: table-top arms, industrial arms and humanoids. - learned directly from human manipulation data - no teleop/robot data - close to human-level dexterity and efficiency - multi-robot collab

Super excited to introduce Fire3D, which turns an unsegmented image or casual RGB video into a complete, textured, simulation-ready 3D scene in under a minute. No supplied object poses or segmentations. Project Page: xiahongchi.github.io/Fire3D/

The Waabi team is headed to Vienna for ICRA, the International Conference on Robotics and Automation 2026, June 1–5! We're presenting four research papers and celebrating a major milestone — our very own @JurvandenBerg will be honored with the @IEEEorg Most Influential Paper Award. Come find us and explore the future of Physical AI and autonomous driving: waabi.ai/icra-2026

The latent-vs-pixel debate misses the point. GPT Image 2 shows what users notice: pixel-level fidelity. Latent models show what scales: compact semantic structure. We connect them by replacing VAE/RAE decoders with a Pixel Diffusion Decoder. Code and Model available: research.nvidia.com/labs/sil… 🧵(1/N)

Awesome. NVIDIA dropped PiD - fast high-res latent decoding via pixel diffusion! - replace VAE - 4/8x upsampling - 2k decoding in <1s on RTX 5090 - works with FLUX.1/SD3/Z - rapid generation previews sharper details, much lower hardware lag compared to standard methods. research.nvidia.com/labs/sil…

We scaled up Lyra to generate explorable 3D worlds! 🚀 Introducing Lyra 2.0 — turning a single image into a 3D world you can walk through, look back, and even drop a robot into 🤖 Code and Model available today! 🌐 Website: research.nvidia.com/labs/sil… (1/N)

At #GTC26 we are showing AlpaDreams — generative, interactive closed-loop simulation for autonomous driving. World models generate environments in real time, policies or drivers interact, and the simulator updates the world. If you are at GTC, come by the booth and try it out!

This week at @NVIDIAGTC we're presenting AlpaDreams: a generative world model for driving simulation. Compared to standard video models, AlpaDreams is autoregressive, enabling updating the conditioning (simple bounding box world) in closed loop, and multiview-consistent.

A new generation in AV simulation is here! We are announcing AlpaDreams, a real time interactive generative world model for AV simualtion! Just a year ago it took minutes to generate a few seconds of video, today it is real time and interactive! research.nvidia.com/labs/sil…

Here we introduce SAGE: Scalable Agentic 3D Scene Generation for Embodied AI, which can generate sim-ready 3D scenes with agents following user demands at scale, ready for robotic action generation. Paper, code, and SAGE-10k dataset are all released! nvlabs.github.io/sage/

Awesome to see @Waymo launching a new model! 🎉 We worked on a similar demo over a year ago — the code and checkpoints are already open-sourced here: 🔗 research.nvidia.com/labs/tor… Love seeing continued momentum in autonomous driving research 🚗✨

We are building a brain to power the entire Physical AI ecosystem. Our Physical AI Platform combines a verifiable end-to-end AI model capable of reasoning with the world's most advanced neural simulator. For the first time in the industry, one AI brain powers multiple applications—trucks and robotaxis—where progress in one directly improves the other. Proven in autonomous trucking. Now expanding to robotaxis. Built to scale across form factors, geographies, and environments. Learn more: waabi.ai/insights/waabi-secu…

Physical AI's moment is here, and self-driving is the first manifestation of Physical AI that will scale. We have built a Physical AI Platform that enables true scale and generalizes across form factors, geographies, and environments. It is proven in autonomous trucking and we are now expanding to robotaxis. Our Platform combines a verifiable end-to-end AI model capable of reasoning with the world's most advanced neural simulator—powering the @Waabi_ai Driver. This approach enables, for the first time in the industry, a shared brain across both applications, meaning the same AI model will drive both trucks and robotaxis. This is important as any progress will directly benefit both verticals. Furthermore, our capabilities already cover highways and surface streets, which has enabled our Direct to Customer model in trucking, providing a big product advantage with the rest of the industry. This also means that we have already all the core capabilities needed for robotaxis, allowing us to enter very fast and seamlessly this new market. To achieve this, we have raised $1B USD of new capital, the largest fundraise in Canadian history. This capital will be used to accelerate commercial progress in autonomous trucks and to fuel our expansion into robotaxis. This funding includes an oversubscribed $750M Series C led by @khoslaventures Ventures and @G2VPLLC , as well as additional capital from @Uber tied to robotaxi development. We are thrilled to work with Uber to deploy robotaxis powered by the Waabi Driver on the Uber platform, the largest ridesharing network globally. Through this strategic partnership, we will deploy 25,000 or more Waabi Driver-powered robotaxis, substantially accelerating the adoption of robotaxis at scale. I am incredibly grateful to our team for making this vision a reality, to our investors, partners and customers for believing in what we are building together. Read more: waabi.ai/insights/waabi-secu…

We have built a Physical AI Platform that enables true scale and generalizes across form factors, geographies, and environments. It is proven in autonomous trucking and we are now expanding to robotaxis. Learn about what we are building: waabi.ai/insights/waabi-secu…

Today, we are thrilled to announce that we have raised $1B USD of new capital to accelerate commercial progress in autonomous trucks and expand to robotaxis! For the first time in the industry, this enables one shared brain to power both applications—trucks and robotaxis. This means any progress in one vertical directly improves the other. This funding includes an oversubscribed $750M Series C round led by @khoslaventures and @G2VPLLC, as well as additional capital from @Uber tied to robotaxi development. We are excited to work with Uber to deploy 25,000 or more Waabi Driver-powered robotaxis on the Uber platform, substantially accelerating the adoption of robotaxis at scale. Thank you to our incredible team, investors, partners, and customers for helping us achieve this milestone. Here's to pioneering the future of Physical AI, together.

Bring new robot testing environments to life with @theworldlabs and Isaac Sim. 🤖 If you can describe a world 🌎, you can start testing in it the same day. Learn how to: 1. Export scenes from World Labs' Marble as Gaussian splats 2. Convert to USD using @nvidiaomniverse NuRec 3. Import into NVIDIA Isaac Sim 4. Add a robot and run the simulation Read the guide ➡️ nvda.ws/3N4Yf5K #SIGGRAPHAsia2025

Flux4D: Flow-based Unsupervised 4D Reconstruction Abstract (excerpt): Flux4D is a simple and scalable framework for 4D reconstruction of large-scale dynamic scenes. It directly predicts 3D Gaussians and their motion dynamics to reconstruct sensor observations in a fully unsupervised manner. By adopting only photometric losses and enforcing an "as static as possible" regularization, Flux4D learns to decompose dynamic elements directly from raw data without requiring pre-trained supervised models or foundational priors. This is achieved simply by training across many scenes. Our approach enables efficient reconstruction of dynamic scenes within seconds, scales effectively to large datasets, and generalizes well to unseen environments, including rare and unknown objects. Experiments on outdoor driving datasets show that Flux4D significantly outperforms existing methods in scalability, generalization, and reconstruction quality.

Flux4D: Flow-based Unsupervised 4D Reconstruction @wangjksjtu, @hungdche, @ycchen918, @yangze0930, @lily_goli, Sivabalan Manivasagam, @RaquelUrtasun tl;dr: photometric losses and simple static-preference prior arxiv.org/abs/2512.03210