NVIDIA reveals how robots learn precision tasks through simulation. Robots are getting better at the delicate work. At this week's Conference on Robot Learning in Austin, Texas, NVIDIA is pulling back the curtain on how artificial intelligence and simulation are transforming the way robots learn to handle precise tasks - from aligning gears to connecting cables and assembling intricate components. For years, simulation has been the training ground where robots perfected their movement skills. Now that same approach is unlocking something harder: manipulation tasks requiring finesse and precision. This shift from basic locomotion to complex hand-like work represents a fundamental change in robotics, driven by new technologies that combine world action models, advanced physics simulation, and autonomous AI agents that can turn individual skills into coordinated robot programs. The announcement comes as the robotics industry grapples with a core challenge: getting robots deployment-ready for real manufacturing and logistics work. NVIDIA's new toolkit addresses this directly through several major releases and partnerships with companies like Ambi Robotics, Bosch Research, Siemens, and others. **GR00T 2 Changes What's Possible** At the heart of NVIDIA's announcement is GR00T 2, an open frontier world action model built on NVIDIA Cosmos. This isn't just incremental progress. Trained on 100,000 hours of diverse human and robot data, GR00T 2 demonstrates capabilities that would have seemed science-fiction just months ago. The model completed 40 out of 50 evaluated tasks without any task-specific demonstrations - meaning it showed up ready to work. When given just one demonstration per task, it achieved 71% average success across those 50 tasks, jumping to 95% success with five demonstrations. More impressively, it can learn from human demonstrations and transfer those skills between different robot bodies without ever practicing on those specific robots. These aren't theoretical benchmarks. Real companies are already testing the technology. Nexuni is using GR00T 2 for robotic laundry handling, Foxconn is applying it to factory automation for server manufacturing, and Siemens is evaluating it across industrial use cases. MicroAGI has even deployed it for Ethernet cable swapping with a robotic arm. **Simulation Now Handles Complexity** Before robots can work in factories, they need to practice. That's where Isaac Lab 3.0 comes in. The latest version of NVIDIA's open-source robot-learning framework introduces capabilities that weren't possible before, particularly through tighter integration with Newton, an open-source physics engine. Traditional physics simulators struggled with real-world complexity. Now researchers can simulate deformable objects like cables and cloth, along with intricate systems like robotic hands with dozens of interacting joints. That accuracy matters enormously - when a robot trains on something that doesn't behave like reality, those skills don't transfer to actual hardware. The performance improvements are striking. Isaac Lab 3.0 with Newton achieved up to 2.5 times the training throughput compared to the previous version. That means researchers can train thousands of simulated robots in parallel, compressing months of development into weeks. Skild AI took this further by announcing STEEL, a new rigid-body physics solver built on Newton specifically designed to accelerate robot training. Flexion showcased how these tools help train Reflect, its robot-intelligence model that combines reasoning with whole-body manipulation. **Making Robots Teachable** Two NVIDIA research papers landed among just six finalists out of thousands of submissions for CoRL's Best Paper Awards. GRAIL introduces a digital generation pipeline for 3D assets and simulator-ready scenes for robot learning. RoboTTT demonstrates how a robot model can scale visuomotor context three orders of magnitude beyond current state-of-the-art policies without slowing down inference. The University of Washington contributed another breakthrough: Success-Guided Sampling (SGS), a training method that focuses practice on tasks where a robot succeeds about half the time. This approach can scale training beyond a million parallel environments. Proof of concept comes from AlphaNIST, built in Isaac Lab, which trained a single policy to handle all 20 tasks on the NIST Assembly Task Board, manipulating millimeter-scale parts with nothing but yes-or-no success signals. It achieved over 98% training success while using roughly 39 times less GPU time than separate per-task policies. **Agents That Code Robot Programs** Perhaps the most significant announcement involves what the industry is calling "agentic robotics" - a paradigm shift where AI agents create interpretable code for physical systems. UC Berkeley's Graph-as-Policy (GaP) framework uses agents to connect individual robot skills into complete programs, then tests those programs in simulation and refines them based on failures. This keeps programs interpretable for human engineers to review before deployment - a critical requirement for industrial applications. Ken Goldberg, a leading UC Berkeley robotics professor, called agentic robotics "the most exciting technological development since the internet in 1995." Ambi Robotics, Bosch Research, and Siemens are all exploring GaP for industrial automation workflows. **Measuring What Matters** Learning skills is one thing. Performing reliably across changing conditions is another. NVIDIA Isaac Lab-Arena provides a framework for scaled policy evaluation that lets researchers run policies across thousands of variations in parallel, then analyze exactly which conditions cause failures. Two new industrial benchmarks emerged from this work. DexBench, developed by RLWRLD with NVIDIA, tests robot dexterity across 15 tasks spanning retail, logistics, manufacturing and hospitality. VABAR, from UC Berkeley and NVIDIA, evaluates robots across six standardized industrial tasks with difficulty tiers and consistent success metrics. These benchmarks matter because they move evaluation from lab tabletops into real-world workflows. Genesis AI, Mimic Robotics, and other partners are already using them through early access programs. The robotics industry has long promised breakthroughs. What NVIDIA is demonstrating at CoRL represents the technical foundation making those promises real - simulation so accurate it transfers to hardware, AI models that generalize across tasks and robots, and agents that turn skills into deployable programs. The next phase of industrial robotics isn't coming someday. It's already rehearsing in simulation. Original story via source. Disclaimer: This article has been rewritten and is based on a third-party news source. Content is provided for informational purposes only. Please visit the original source for full details. Digital Ground Media does not claim ownership of the underlying story.

China's robot makers have mastered the moves. Now comes the race for brains. Published Nov 01, 2026, 06:09 PM Updated Nov 01, 2026, 08:24 PM (C) Reuters. By Laurie Chen and Eduardo Baptista BEIJING, Nov 2 (Reuters) - After years of building humanoid robots capable of running, dancing and performing backflips, Chinese companies are pouring resources into a different challenge: developing the AI that could make those machines useful across a wider range of jobs. The shift is fuelling an investment boom in embodied AI, a field focused on software that allows robots to understand their surroundings, learn unfamiliar tasks and adapt without having to be programmed for every job. Chinese embodied AI companies raised 93.5 billion yuan ($13 billion) across 322 deals in the first half of 2026, five times the amount a year earlier, according to data research firm ITjuzi. Funding is increasingly reaching specialist developers of robot intelligence alongside manufacturers. As competition drives down hardware prices, robot makers have "concluded that the margin won't come from the hardware but from the brain," said Dermot McGrath, founder of Shanghai-based technology consultancy ZenGen Labs. The emerging sector is opening a new front in US-China tech rivalry. American companies lead in some advanced robot AI systems, while Chinese developers hope their manufacturing strength and access to real-world training data will help them challenge their US competitors, according to Reuters interviews with founders and top executives at five Chinese robotics firms, as well as industry analysts. The country that best combines AI with machines could gain an advantage in automating factories and logistics, potentially reshaping industrial productivity and the global market for robotics technology. -9.02 (-4.22%) Closed · 30/10 · USD +2.52 (+1.23%) After Hours · 18:55:47 "The brain is indeed the weakest link in the complete robotics stack," said Spirit AI co-founder Gao Yang. RACE FOR INTELLIGENCE American startups, including Physical Intelligence, Skild AI and Generalist, are developing software intended to teach robots different skills without extensive retraining. Generalist says its GEN-1.5 model completed around 60% of simple tasks after viewing a demonstration lasting between 3 and 12 seconds. China's X Square has published research showing robots acquiring handling skills from a single human video without retraining the model for each task. "The leading American labs are setting a demanding benchmark, particularly in how quickly robots can acquire and reuse skills," said robotics consultant Georg Stieler, who considers China's strongest startups credible competitors. While US developers benefit from advanced AI research and greater access to powerful computing chips, Chinese companies can work closely with robot manufacturers and industrial customers to test and improve their systems. Specialist Chinese startups, including X Square, Galaxea and Spirit AI, have each reached valuations of roughly $3 billion, according to company disclosures. "We want robots to shift from being driven by code to being driven by common sense," said X Square co-founder and COO Yang Qian. Established manufacturers are investing too. Hong Kong-listed UBTech advertised a chief scientist position overseeing AI model development with compensation of up to 124 million yuan ($18 million), including equity and benefits. Hangzhou-based Unitree, one of the world's largest humanoid robot makers by shipments, has about 30 researchers developing AI that predicts how objects behave, according to a person briefed by the company. Unitree did not respond to a request for comment. THE TRAINING RACE Unlike chatbots, which learn from vast quantities of internet text, robots need examples of physical interactions, from gripping objects to handling flexible wires. Chinese companies are pursuing different approaches to collecting that information, with some employing people to demonstrate tasks, others using computer simulations, and some combining both. The field remains crowded, with no company establishing itself as the clear leader. Spirit AI has hired around 1,000 people to record everyday and industrial activities, Gao said. ACE Robotics uses wearable sensors to capture workers' movements and grip forces in stores and hotels. Ant Group's robotics unit Robbyant said it trained a model using 160,000 hours of material, including 50,000 hours of human-guided robot movements. It also trained software on 20 robot configurations from 17 brands, aiming to use the same AI across different machines. DexForce is taking another approach, using computer-generated environments to teach robots. Chief operating officer Nicole Yang said the company can prepare some industrial tasks for deployment within 6 to 8 hours and reported success rates above 99.9% in specified precision operations. But simulations cannot perfectly reproduce physical interactions, said Jerry Wang of spatial intelligence company Manycore Tech, which designs simulated environments for AI training. "Data quality is more important than data diversity, and more important than simply collecting data at massive scale," X Square's Yang said. Chinese robot firms also face additional headwinds, including shortages of advanced Nvidia chips due to US export controls, and a squeeze on IPO funding amid regulatory concerns over inflated valuations and overcrowding. As robots become more powerful, the shortage of advanced chips will become a serious problem for China's embodied AI industry, said Zhu Mingxuan, who leads development of JoyIn Technology's Aether robot AI. FROM TRAINING TO WORK JoyIn and other developers are already testing whether their software can enable robots to tackle unfamiliar problems. Zhu described an experiment in which a robot repeatedly failed to place headphones in a cupboard beyond its reach. Then engineers showed it videos of how people managed the task, and the robot pushed a box towards the cupboard and tried stepping onto it, Zhu said. "Every time it was exploring a different behaviour," she said. Real-world deployments in China leave no room for improvisation. Battery maker CATL said Spirit AI robots connect flexible wires during battery-pack testing at its Zhongzhou plant, reporting insertion success above 99%. Robbyant said its robots handle orders in two Shanghai pharmacies, with staff checking medicines before dispatch. But the longer-term prize is software capable of learning new jobs without extensive adaptation. Success will depend how the same AI performs across different customers and workplaces and that will require companies to train increasingly powerful models with advanced chips. Chinese companies are trying to reduce their computing requirements through more efficient models, synthetic data and simulations, while domestic chipmakers, including Huawei, are developing alternatives to Nvidia hardware. It remains unclear how far those efforts can offset the shortage. 'CHATGPT MOMENT' Unitree founder Wang Xingxing said last month the company had tested OpenAI's GPT-6 on robots with encouraging results. He said GPT-7, if released next year, could bring robotics its own "ChatGPT moment" - a step-change in capability akin to ChatGPT's 2022 launch, which turned generative AI into a mainstream consumer and business technology. Wang said he would consider that threshold reached when robots were able to complete 80% of tasks in 80% of unfamiliar environments from spoken instructions. But consultant Stieler said a breakthrough in demonstrations would not necessarily mean robots were ready for broad commercial use. "If the question is when a model first handles a broad range of unfamiliar tasks in demonstrations, the optimists may well be right," he said. "If the question is when a customer can count on it, I would add years, not months." See the trade on NVDA, but can't pull the trigger? Most traders can read a chart. The hard part is the moment: entry window open, pattern forming, and you're still waiting for more confirmation. That's the conviction gap - and its chart analysis closes it. Unlike other AIs that just read data, its Vision AI literally "sees" your charts and hands you a complete trading plan: entry, stop-loss, and profit target in under 60 seconds. Know exactly what to do next, every time.