Practice makes precise: at CoRL, NVIDIA and partners reveal how robots rehearse delicate work in Simulation. 09/11/2026 Simulation, which has long been the proving ground for robots learning to walk, roll and hop, is now helping researchers teach robots tasks that require a finer touch: aligning gears, connecting cables and assembling small components. The shift from locomotion to manipulation is taking center stage at this week's Conference on Robot Learning (CoRL) in Austin, Texas, alongside the rise of agentic robotics - a new paradigm that integrates frontier agents to create interpretable code for physical robotics systems. At CoRL, NVIDIA is introducing new open robotics technologies, including world action models, simulation tools, benchmarks and agentic workflows. Two NVIDIA papers are among six finalists out of thousands of submissions for the Best Paper Awards at CoRL: GRAIL (Generating Humanoid Loco-Manipulation from World Models), which presents a digital generation pipeline for 3D assets and simulator-ready scenes for robot learning. RoboTTT: Context Scaling for Robot Policies, which introduces a robot model and training recipe that scales visuomotor context three orders of magnitude beyond state-of-the-art policies without growing inference latency. These advancements help researchers train robots on complex tasks, evaluate their performance across changing conditions and use failures to guide further learning - building the precision, adaptability and autonomy needed to move from lab demonstrations to industrial deployment. NVIDIA is working with partners like Ambi Robotics, Bosch Research, Dexmate, Dyna Robotics, FieldAI, Flexion, Noitom Robotics, Siemens, Skild AI, RLWRLD, UC Berkeley, Walden Robotics and the University of Washington to bring these advancements to the broader robotics community. World Action Models Enable Next Level of Robot Autonomy The next level of robot autonomy requires robots to perceive their surroundings, anticipate how their actions will change the environment and execute complex, multistep tasks with a high degree of precision. World action models enable exactly that. At CoRL, NVIDIA is introducing GR00T 2, an open frontier world action model built on NVIDIA Cosmos. Trained on diverse robot and human data, including 100,000 hours of NVIDIA open source human egocentric data, the model enables robots to learn dexterous manipulation skills. Key features include: Out-of-the-box performance: Completed 40 of 50 evaluated tasks at least once without task-specific demonstrations. Skill transfer: Learns from human demonstrations and transfers skills between different robot bodies without robot demonstrations Generalization with little data: Achieved 71% average success across 50 tasks with just one demonstration per task, increasing to 95% with five. Agent control: AI agents can use GR00T 2's application programming interface and supporting tools to run skills in sequence and recover from errors. GR00T 2 currently ranks among the top three models across all vision language action and world action model categories for the RoboCasa365, RoboLab, RoboDojo and MolmoSpaces benchmarks, with multiple leaderboard wins. Nexuni is testing it for robotic laundry handling, Foxconn is applying GR00T 2 to factory automation for server manufacturing and Siemens is evaluating it across industrial use cases. MicroAGI is using GR00T 2, post-trained through reinforcement learning, to perform Ethernet cable swapping with a YAM robotic arm. Its agentic robotics approach automates the creation and calibration of simulation environments using Isaac Lab 3.0 and Newton, where the policy is trained. Training capable robots requires diverse data. Open NVIDIA workflows and models equip researchers with better tools for data collection and annotation: For researchers building robot-training datasets, NVIDIA's new open Video to Data workflow turns human video into simulated robot demonstrations, with a challenge inviting researchers to advance it. The NVIDIA EgoHand open model - being demoed at CoRL and with over 40,000 Hugging Face downloads in the two weeks since launch - reconstructs 3D hand motion, trained with data from Noitom Robotics. Where the Training and Scaling Happens - in Simulation A pretrained model gives robots a core set of skills. Simulation lets researchers train, refine and evaluate skills at scale before deploying them on physical robots. NVIDIA Isaac Lab, an open source robot-learning framework, provides the foundation for that work. Its latest release, Isaac Lab 3.0, allows researchers to use agents to quickly set up training environments with simulation-ready assets, switch between NVIDIA PhysX and Newton physics solvers, choose renderers for perception-based training and apply a new reinforcement post-training library to foundation models like GR00T. Accurate physics is critical to transferring learned skills from simulation to the real robot. Tighter integration with Newton, an open source physics engine, now allows Isaac Lab to model deformable objects such as cables and cloth, along with complex articulated systems such as robotic hands with many interacting joints. Integration with Newton simulates how a strawberry deforms under a robot's grip. Together, Isaac Lab and Newton combine accurate physics with scale to enable researchers to train thousands of simulated robots in parallel. In a vision-based dexterous manipulation benchmark, Isaac Lab 3.0 with Newton achieved up to 2.5x the training throughput of Isaac Lab 2.3 with PhysX. Isaac Lab supports robot simulation with Newton and NVIDIA PhysX. Developers are extending Newton'

Agentic Robotics solves an industrial production problem. By Ambi Robotics 09.17.2026 Using a new Graph-as-Policy-based Agentic Robotics harness within AmbiOS, AI agents solved in 10 hours a package-placement problem that would have taken engineers several weeks to address. The solution is now deployed across 30% of Ambi Robotics' U.S. fleet, including systems operating for a Fortune 50 package shipping company. To the best of its knowledge, this is the first time Agentic Robotics has solved a real production problem. Improving performance on commercially-deployed robots requires significant human effort. Engineers must review key data offline (e.g. videos, images, sensor data, and logs), hypothesize root causes, and test updates. This process is slow and requires significant domain knowledge. "Agentic Robotics" offers an exciting new approach. The idea is to use AI coding agents to automatically analyze data, identify problems, and solve them to improve robot performance. This combines rapid advances in AI coding with classical model-based robotics modules and learned model-free policies. This requires agentic robotics "harnesses" to guide AI coding systems to solve the real robotics problems. Several new research papers, including Code-as-Policy (CaP) [1, 2] and Graph-as-Policy (GaP) [3], show surprising promise for rapidly generating reliable robot control systems with agentic harnesses that constrain agents to use libraries of relevant robot skills. Building on these ideas, Ambi Robotics developed the AmbiOS Agentic Robotics Harness. It builds on GaP to update computation graphs that are composed of existing robot skills. The harness gives agents access to two key pieces of infrastructure. First, agents access datasets of real robot events from the cloud that can be filtered by a set of known success and failure modes. Second, agents use a simulation environment to replay these events with new self-generated behaviors and test graph edits, evaluating the outcomes in the simulation. The harness constrains how agents work, ensuring that the final result can be executed without compromising safety checks and error handling. Figure 1. AmbiOS Agentic Robotics harness architecture for performance improvements on real production robots. Task. The AmbiSort robot picks, scans, and places random packages with a robot arm and a gantry receives the package and sorts it into a bag based on data like the zip code. The target bags have a fixed opening size through which items must fit to fall into the bag without jamming. Ambi Robotics applied the AmbiOS Agentic Robotics Harness to a problem where subtle edge cases impact critical operational metrics: throughput, sort accuracy, and uptime. Millimeter-level inaccuracies can have a significant effect on throughput. Packages that are on the boundary of fitting into the bag opening are challenging because the robot arm cannot estimate item dimensions and pose with perfect precision due to sensor noise and item deformation. Furthermore, the robot prefers placing items with the short side going into the bag to maximize space utilization. As a result, these edge case items are too wide to fit through the bag opening, forcing the system to reject and retry packages, losing valuable time. Ambi Robotics estimated that this issue was reducing capacity by approximately 31k sorts per year per robot for a typical facility running 12 hours a day, 6 days a week. Fixing this issue with hand-tuning would take several weeks of engineering time. Figure 2. Example of a retry due to the item not fitting into the sack. The item is drooping slightly when lifted, leading the robot to underestimate the item width and place it in a poor orientation. The gantry drops the item back into the input bin to retry, costing valuable time. Ambi Robotics used the AmbiOS Agentic Robotics Harness with Anthropic's Claude models including both Sonnet 4.6 and Opus 4.8. The agents analyzed images and data from 500 events over the previous week of production, 50% of which were the known failure mode and 50% which were selected at random to avoid regressions. Within 10 hours, the agents generated and evaluated three solution hypotheses. Results. Ambi Robotics deployed these three hypotheses into real production A/B tests in customer warehouses. Each solution was compared to a baseline software version. Here are the results after 10 days: | Hypothesis | Estimated Change in Sorts per Year per Robot | | H1 | −374 | | H2 | −1,497 | | H3 | +15,725 | Table 1. Comparison of throughput and estimated packages per year across three A/B test solutions deployed to real production robots. For H3, the agent noticed that for large, dangling bags and paper mailers, the placement plan needs to carefully trade off orienting the item to fit in the destination bag and avoiding placements where the item could get stuck in grates on the placement platform. It optimized this tradeoff and decided to route more packages through a subgraph for planning dangling item placements because it results in a more predictable final orientation. In addition, the agent changed the flat item subgraph to only optimize for the final orientation of the item because it is easier to predict the final resting position of flat items than for bags. In real production A/B tests, H3 increased throughput by 4.2 packages per hour. Over 12 hours per day and 6 days a week (standard operation), this corresponds to over 15k packages per year per robot. Figure 3. Placement behavior before and after S3, for a set of near-identical packages in production. Select an example to load it, then drag a view to rotate both. Conclusion. To the best of its knowledge, this is the first time that Agentic Robotics has solved a real production problem. The solution runs on 30% of the Ambi Robotics fleet sorting real packages with 24x7 availability. Ambi Robotics is now applying the AmbiOS Agentic Robotics Harness to a number of other production problems, accelerating performance improvements. Want to scale physical AI in real world deployments? Ambi Robotics is hiring! References. [1] Liang, Jacky, et al. "Code as policies: Language model programs for embodied control." 2023 IEEE International conference on robotics and automation (ICRA). IEEE, 2023. [2] Fu, Letian, et al. "CaP-X: A framework for benchmarking and improving coding agents for robot manipulation." arXiv preprint arXiv:2603.22435 (2026). International Conference on Machine Learning (ICML), July 2026. [3] Chen, Kaiyuan, et al. "GaP: A Graph-as-Policy Multi-Agent Self-Learning Harness For Variational Automation Tasks." arXiv preprint arXiv:2607.05369. To Appear: Conference on Robot Learning (CoRL), Nov 2026.