Check out An-Chieh Cheng's new work on providing a scalable way for VLM to have the sense of 3D and perform 3D spatial reasoning in any scenes: https:// anjiecheng.me/sr3d

A Vision-Language-Action AI Model applied to Industrial Robotics Instead of building separate models for each robot or task, Sereact has created Cortex, a foundation model that can control different robot types (arms, mobile manipulators, humanoids) and adapt to new environments. What makes Cortex interesting? ✅ One model, many robots: arms, mobile manipulators, humanoids ✅ Day-one value: production-ready in hours, no retraining or per-site code ✅ Edge-case resilience: robust to slips, drops, glare, and clutter ✅ Language as interface: operators can set goals in plain English or German ✅ Fleet learning loop: every deployment makes the model smarter ✅ Already deployed across 100+ industrial systems Cortex is one of the first solution to deploy VLA AI model on industrial robotics applications. Already adopted by big actors of the industry, it opens various possibilities for the future of industrial robotics and automation. 🔗 Check the project page to learn more about Cortex! How do you think VLA and LLM will impact the Industrial Robotics industry? Let's connect and share robotics news 🔽 #Robotics #AI

MorphLab — Towards Self-Evolving Robotic Bodies and Minds https://lnkd.in/gHyrpC6E Co‑design robot morphology and control policy in simulation: evolving finger geometries (via genotypes) for manipulation tasks. First, I've combined dense 3D tactile sensing in simulation with tactile-aware reinforcement learning policies — enabling a robot to not only act, but feel. Then I built a full morphology evolution framework that allows the robot’s fingertip shape and material to evolve over time — driven purely by performance on manipulation tasks, and through evolution (mutation, crossover, and selection). The result? * More dexterous fingertips * More robust manipulation This is a compelling demonstration of how touch-guided evolution can unlock better designs than we might imagine by hand. #EmbodiedAI #Robotics #EvolutionaryAlgorithms #IsaacSim #ReinforcementLearning

Sure, watching robot arms gracefully open and close blinds is satisfying, but the real MVP here? The incredibly detailed 3D simulation that made this possible. Think of us as the movie set designers of the robotics world. While everyone's watching the actors (robots) perform, we're the ones who built the hyper-realistic stage where they learned their lines. Our photorealistic, physically-accurate 3D assets are so good that robots graduate from our virtual "training academy" and immediately nail their real-world debut. No awkward "I thought that was supposed to work differently" moments. Because let's be honest - nobody wants their expensive robot to have an existential crisis when it encounters its first real curtain. 🎯 Ready to build your robot's dream training ground? Explore our simulation-ready 3D assets at simready.com or connect with us https://lnkd.in/eVv3e2Kk to design your own custom training academy. Your robots will thank you 😆 (once they learn how). #Robotics #EmbodiedAI #Simulation #IsaacSim #IsaacLab #GR00T #VLA #Teleoperation #LeRobot #SimreadyAssets #RLTraining #3DAssets

WATCH : Boston Dynamics’ Atlas just got a new mind. For years, humanoid robots were trapped by their code. Engineers had to handcraft every routine, from walking to lifting. Now, Atlas is stepping into a new era powered by Large Behaviour Models (LBMs). What’s new Whole-body intelligence: Atlas can squat, bend, and coordinate arms and legs together, treating its limbs as interchangeable manipulators. Language + sensors → action: LBMs digest camera input, proprioception data, and short natural-language instructions, then output fluid, real-time movements. Speed and efficiency: Atlas learns new tasks 3–5x faster than traditional training and executes them up to 2x faster than human demonstrations. Adaptability: Mid-task, Atlas can adjust. If a lid closes or an object shifts, it recovers without new coding. Why it matters Boston Dynamics and Toyota Research Institute are proving that robots do not need hard-coded skills anymore. With LBMs, a single generalist policy can handle dozens of tasks, from lifting a box to folding cloth. This is the pivot: from robots as narrow specialists to robots as general-purpose workers. The equivalent of what foundation models did for language, LBMs are doing for robotics. We are now watching humanoids evolve from choreographed machines to adaptable teammates, closer to real-world deployment than ever before. #AI #Robotics #Robots #BostonDynamics #LBM

Training physical robots just got a virtual upgrade. 🧠🔧 MIT CSAIL and the Toyota Research Institute have developed a groundbreaking technique called steerable scene generation—revolutionizing how robots learn tasks like placing dishes or organizing kitchens. Instead of relying on costly and inconsistent real-world data, this method uses highly realistic, physics-aware 3D environments for robot training. With tools like Monte Carlo Tree Search and reinforcement learning, robots can now learn from tailored virtual setups, improving performance while reducing cost and time. Explore how this fusion of AI, robotics, and simulation is transforming the future of automation: https://lnkd.in/dAQSQPCt #ArtificialIntelligence #Robotics #GenerativeAI #MITCSAIL #ToyotaResearchInstitute #RobotLearning #AIResearch #3DSimulation #Automation #FutureTech

Virtual training for robots is leveling up—MIT CSAIL and Toyota Research Institute are using generative AI and simulation to teach robots complex tasks in physics-aware 3D environments. Smarter learning, lower costs, faster deployment. #Robotics #GenerativeAI

A major leap for robot learning: steerable scene generation enables training in physics-aware, photorealistic simulations—cutting costs and accelerating development. Exciting progress at the intersection of AI and robotics. #RobotLearning #3DSimulation

Training robots in photorealistic, physics-aware virtual environments is a game-changer. MIT CSAIL and Toyota Research Institute are redefining robot learning—faster, cheaper, and smarter. A major step toward scalable, real-world autonomy. #RobotLearning #3DSimulation

Hanging objects on moving hooks — a common task in manufacturing (like coating) — is surprisingly hard for robots. Why? ✅ The line is constantly moving, so the robot must adapt in real time. ✅ Hooks are thin, objects are reflective — making 3D vision unreliable. With SigmaKit, this challenge becomes routine: ⚙️ Our real-time inference lets robots track and adapt to moving hooks. 👁️ Our vision system accurately detects both hooks and reflective objects. Out of hundreds of demos, we’ve never missed a hook. Not once. #robotics #automation #manufacturing #AI #computerVision #SigmaKit