🍾🍲 Saturday Robotics x IROS 2026 — Robotics Research Night 👉🏻 https://luma.com/tzbw7n61 We’re bringing a high-signal evening of robotics research to Pittsburgh on September 28. After a full day at IROS, we’ll bring together researchers, engineers, founders, students, and investors for technical discussions, networking, and a series of ~10-minute lightning talks. Tentative preview of the current lineup: 🤖 1. PAC-MAN: Perception-Aware CBF-RL for Whole-Body Safety in Humanoid Dodgeball Gary Yang @lzyang2000 (@Caltech) Perception-aware reinforcement learning + Control Barrier Functions for whole-body humanoid safety. Demonstrated on a Unitree G1, with 19/20 successful dodges and zero falls in real-world experiments. 🧠 2. How In-Context Learning Is Reshaping Robot Learning Data at Scale AaronLi (@RhodaAI) Exploring how in-context learning can change the way we think about robot learning data, scaling, and generalization. 🧪 3. X2Real: An eXtensive Simulation Benchmark for Real-World Generalist Policies Liangwang Ruan (@XSquareRobot) A new simulation benchmark built around faithfulness, diversity, and fairness, with 44 hierarchical long-horizon tasks across 10 capability dimensions and a reported 0.84 simulation-to-real correlation. 🦾 4. Rethinking Generalist Robotic Manipulation: Architecture, Data and Inference for Real-World Deployment Peiyan Li (Chinese Academy of Sciences, @CAS__Science) 3D VLA architectures, memory augmentation, ego/UMI human priors, large-scale robot pretraining, and inference-time contextual learning for deployable generalist manipulation. 🎯 5. HiRE: Hindsight Reward Editing for Policy Finetuning Haoyi Niu @t641769919 (@UCBerkeley) Accepted at CoRL 2026. A training-free approach to reward editing that uses successful and failed trajectories to identify “trap states” and provide denser, control-aware feedback for RL. 🔥 6. Lightning Talk — Open Slot We’re opening one additional slot for a technically deep research talk, new project, frontier paper, demo, open problem, or startup technical insight. 10 minutes. A few slides. One sharp technical idea. No fluff. Topics include World Models, Physical AI, Humanoids, VLAs, Robot Foundation Models, Manipulation, RL, Simulation & Sim-to-Real, Spatial Intelligence, Computer Vision, and Embodied AI. 📍 Pittsburgh 📅 September 28, 2026 🕠 5:30–9:30 PM 🍾 Networking + Technical Talks + Research Discussion 📩 junfanzhu98@gmail.com See you in Pittsburgh. 🤖 #IROS2026 #Robotics #PhysicalAI #RobotLearning #WorldModels #HumanoidRobotics #VLA #EmbodiedAI #RobotFoundationModels

We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics. The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra. The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.

LLMs with scaffolds have lagged on text-to-SQL, a task that relies on human judgment. By folding expert judgment into every part of RLVR on Tinker, @maxYuxuanZhu and @ddkang (UIUC and Bridgwater) trained the first text-to-SQL model to beat the human mark. https://thinkingmachines.ai/news/putting-task-expertise-into-rl

🚀 What if video generators could build on representations that already understand the visual world? We are excited to introduce V-RAE: Rethinking Video Latent Spaces for Generation. Recent progress in image generation has begun to move beyond conventional VAE latents, exploring both direct pixel-space and representation-based approaches. Video generation, however, still depends heavily on latent compression, as the scale and redundancy of spatiotemporal data make direct modeling prohibitively expensive. However, most video VAEs are optimized for pixel reconstruction, and a latent space that reconstructs well is not necessarily easy to generate. V-RAE takes a different approach: it directly uses representations from frozen vision foundation models as the generative latent space, rather than as auxiliary supervision. We study DINOv3, SigLIP2, EUPE, and V-JEPA 2.1. A lightweight temporal attention pooling module compresses their dense features by 4×, followed by a spatiotemporal Transformer decoder. Under matched generation backbones, latent budgets, and training settings, V-RAE achieves: 🏆 2.13 rFVD on Kinetics-600 🎬 117.86 gFVD on UCF101 and 19.16 gFVD on Kinetics-600 ⚡ Up to 6× faster convergence than VAE-based latent spaces 🧠 90.92% semantic probing accuracy on UCF101 🌍 Better future prediction on Cityscapes, reducing gFVD from 144.47 to 111.36 Our experiments also reveal a broader finding: Good Reconstruction ≠ Good Generation. During generation, predicted latents inevitably deviate from real encoding trajectories. If the latent space is not sufficiently smooth, small errors can be amplified into visible artifacts. We therefore introduce tFVD to evaluate temporal smoothness and robustness to latent prediction errors. It correlates much more strongly with downstream generation quality, reaching 0.919 on Kinetics-600. The takeaway: A latent space is not merely where videos are compressed—it determines what the generator must learn. When semantics and temporal structure are already organized in the representation, generation becomes easier to learn. Representation first. Generation follows. Many thanks to my mentors, @ScottNLP and @SQWu_Tori, for their continuous guidance and support. I am also deeply grateful to @sainingxie for his valuable guidance and invaluable feedback, which greatly helped shape V-RAE. 🙏 Hi @_akhaliq, we would truly appreciate your help in sharing V-RAE with the broader AI research community. Thank you! 🙏 📄 Paper: https://arxiv.org/abs/2608.13556 💻 Code: https://github.com/V-RAE/V-RAE 🤗 Models: https://huggingface.co/Guomh0707/V-RAE-Models 🌐 Project: https://v-rae.github.io #VideoGeneration #GenerativeAI #ComputerVision #WorldModels #RepresentationLearning #RAE