# What is true on Runtime? > Markdown twin of https://roboterminal.app/topics/runtime · https://roboterminal.app/topics/runtime.md ## On this layer - org — [NVIDIA](https://roboterminal.app/companies/nvidia) (9) - org — [Skild AI](https://roboterminal.app/companies/skild) (2) - org — ADLINK (1) - org — Apple (1) - org — Applied Intuition (1) - org — articulated assets (1) - org — Astribot (1) - org — Axis Robotics (1) - org — [Boston Dynamics](https://roboterminal.app/companies/boston-dynamics) (1) - org — Continuum Labs (1) - org — development timeline of Moby. (1) - org — development timeline reduction. (1) - org — examples for adapting to new robotic embodiments. (1) - org — Fudan University (1) - org — [Google DeepMind](https://roboterminal.app/companies/google) (1) - org — [Hugging Face](https://roboterminal.app/companies/hugging-face) (1) - org — Lightwheel (1) - org — LimX Dynamics (1) - org — LM Studio (1) - org — manipulation tasks (1) - org — Moby deployment in a semiconductor facility. (1) - org — Noble Machines (1) - org — Ollama (1) - org — Perplexity (1) - org — post-training duration (1) - org — SAGE-10K dataset (1) - org — Scale AI (1) - org — Scale AI data delivery (1) - org — Schaeffler (1) - org — Seeed (1) - org — Solomon (1) - org — Stretch (1) - org — success rate of the post-trained policy. (1) - org — tasks available in RoboLab. (1) - org — The University of Hong Kong (1) - org — Training Data (1) - org — training hours for LDA-1B model (1) - org — training needed for current VLA models to match S1's accuracy (1) - org — TranscEngram (1) - org — TRON 2 (1) - org — validated run uses 64 nodes of 4× GB200 for 60K iterations, roughly 68 hours (17.4K GB200-hours). (1) - org — Zhejiang University (1) - org — ZiNovaLabs (1) ## Who captures - NVIDIA (9) - Skild AI (2) - Continuum Labs (1) - Stretch (1) - Astribot (1) - @smsehy (1) - Perplexity (1) - https://x.com/VaderResearch (1) - ZiNovaLabs (1) - The University of Hong Kong (1) - https://x.com/RoboPapers (1) - Axis Robotics (1) - Google DeepMind (1) ## Sourced numbers - **manipulation tasks:** 120 units - _"across 120 language-conditioned manipulation tasks"_ - Source: [@NVIDIARobotics on X](https://x.com/NVIDIARobotics/status/2095179991650644018) - **TRON 2:** 1 units - _"TRON 2 serves as a modular and extensible embodied robotic platform for multi-tool, multi-step tasks across large workspaces."_ - Source: [LimX Dynamics on X: "Together with ZINOVA's Tool Intelligence, TRON 2 takes on increasingly complex construction workflows. @ZiNovaLabs builds on TRON 2 to explore an innovative robotic configuration for construction, demonstrating key tasks in a scaled-down tilt-up construction workflow, includi… / X](https://x.com/chris_j_paxton/status/2095114381818253768) - **Moby deployment in a semiconductor facility.:** 3 units - _"bring two Moby3 units to a semiconductor facility for material-handling workflows."_ - Source: [Noble Machines Accelerates Humanoid Robot Development 3X | NVIDIA](https://x.com/NVIDIARobotics/status/2094923298589024698) - **development timeline of Moby.:** $3 - _"Accelerate Moby's development timeline by nearly 3x, from an expected four years to 18 months."_ - Source: [Noble Machines Accelerates Humanoid Robot Development 3X | NVIDIA](https://x.com/NVIDIARobotics/status/2094923298589024698) - **development timeline reduction.:** 18 hrs - _"from an initial estimated timeline of four years with 50 people to 18 months with 15 people."_ - Source: [Noble Machines Accelerates Humanoid Robot Development 3X | NVIDIA](https://x.com/NVIDIARobotics/status/2094923298589024698) - **SAGE-10K dataset:** 10,000 units - _"The SAGE-10K dataset contains 10,000 generated indoor scenes across 50 room types."_ - Source: [How to Train a Cross-Embodiment Robot Navigation Policy with AI Agents | NVIDIA Technical Blog](https://x.com/NVIDIARobotics/status/2093381258243895657) - **post-training duration:** 50 hrs - _"We found that to match the accuracy that S1 can achieve with just one example of prompting, current VLA models would need to be post-trained with 50-100 hours of data collection followed by fine-tuning!"_ - Source: [Skild AI on X: "We found that to match the accuracy that S1 can achieve with just one example of prompting, current VLA models would need to be post-trained with 50-100 hours of data collection followed by fine-tuning!" / X](https://x.com/SkildAI/status/2092300859317330378) - **training needed for current VLA models to match S1's accuracy:** 100 hrs - _"current VLA models would need to be post-trained with 50-100 hours of data collection followed by fine-tuning!"_ - Source: [Skild AI on X: "Introducing S1, our new foundation model that learns from one example. It can be taught 10-minute long tasks that it has never seen before, from one video prompt without any fine-tuning. Watch S1 operate in real-time via in-context learning:" / X](https://x.com/SkildAI/status/2092300842900865389) - **training hours for LDA-1B model:** 30000 hrs - _"LDA-1B s a new foundation model that is trained on 30,000 hours of human and robot interaction data."_ - Source: [RoboPapers on X: "Instead of choosing between training a world model and training a language conditioned robot policy, why not do both? LDA-1B s a new foundation model that is trained on 30,000 hours of human and robot interaction data. Part of the secret is that LDA-1B jointly learns forward … / X](https://x.com/chris_j_paxton/status/2091877313872834800) - **articulated assets:** 27 units - _"TaskGen now includes a library of 27 articulated object families (4 variants each)."_ - Source: [@axisrobotics on X](https://x.com/axisrobotics/status/2091797746114244876) - **validated run uses 64 nodes of 4× GB200 for 60K iterations, roughly 68 hours (17.4K GB200-hours).:** 68 hrs - _"Plan compute accordingly."_ - Source: [Post-Train NVIDIA Cosmos 3 Edge for On-Device Robot Control | NVIDIA Technical Blog](https://x.com/NVIDIARobotics/status/2095179993974260168) - **tasks available in RoboLab.:** 120 units - _"It executes each action chunk in physics and streams rendered observations back for a true closed loop across 120 language-conditioned manipulation tasks."_ - Source: [Post-Train NVIDIA Cosmos 3 Edge for On-Device Robot Control | NVIDIA Technical Blog](https://x.com/NVIDIARobotics/status/2095179993974260168) - **success rate of the post-trained policy.:** 22.9 units - _"In closed-loop RoboLab evaluation, the post-trained Edge policy reaches 22.9% success across 120 language-conditioned manipulation tasks."_ - Source: [Post-Train NVIDIA Cosmos 3 Edge for On-Device Robot Control | NVIDIA Technical Blog](https://x.com/NVIDIARobotics/status/2095179993974260168) - **Scale AI data delivery:** 150000 hrs - _"it reported delivering over 150,000 hours of physical AI data during 2025"_ - Source: [5 Physical AI infrastructure platforms shaping robotics in 2026 - The Robot Report](https://www.therobotreport.com/5-physical-ai-infrastructure-platforms-shaping-robotics-in-2026) - **examples for adapting to new robotic embodiments.:** 200 units - _"we can now adapt to new bi-arm robot embodiments with just a few hours of adaptation time, typically with less than 200 examples."_ - Source: [Gemini Robotics 2 brings whole body intelligence to robots — Google DeepMind](https://x.com/SkillupAi/status/2093896499465945235) - **Training Data:** 20000 hrs - _"N1.7 is pretrained on 20K hours of EgoScale human video data alongside diverse robot demonstrations."_ - Source: [NVIDIA/Isaac-GR00T](https://github.com/NVIDIA/Isaac-GR00T) ## Pulse - [@NVIDIARobotics on X](https://x.com/NVIDIARobotics/status/2095950021636046876) Interested in getting hands-on with Newton? The newly released Newton Learning Path will help you set up a ready-to-run environment, dive into Newton abstractions, and show you how to build your first… - [@continuumlabs_ on X](https://x.com/continuumlabs_/status/2095923406214758560) How do you update 10,000 robots at once? You don’t blindly push an update to all of them. Fleet-scale software updates typically follow a controlled rollout: → Build new software → Test and validate →… - [@NVIDIARobotics on X](https://x.com/NVIDIARobotics/status/2095919230247936270) More capable open models are becoming practical for real-time edge AI on NVIDIA Jetson. ⚡ Choosing the right one is not just about capability. It is about matching the model to the work your agent nee… - [@stretchcloud on X](https://x.com/stretchcloud/status/2095665142730256633) The harness problem is getting its own framework category. Tardigrade ships today: an agent harness built as typed state machine components over an immutable event log. The framing is React for harnes… - [@XRoboHub on X](https://x.com/XRoboHub/status/2095588510950727820) Robots can’t stop while the model thinks. In a high-speed throw, even one pause can kill the momentum. Astribot released SmoothRL for online RL during async inference. S1 keeps moving as the model com… - [@smsehy on X](https://x.com/smsehy/status/2095440999594287518) Writing a custom robotics runtime in Rust eliminates garbage collection latency spikes on the factory floor, but maintaining sub-millimeter trajectory repeatability across varying ambient thermal cond… - [@stretchcloud on X](https://x.com/stretchcloud/status/2095294198886903921) The hybrid compute pattern is getting serious. Perplexity just open-sourced Lily, the local inference engine powering their Mac hybrid compute feature, and the architecture choices are worth understan… - [mete on X: "Buying a quadruped? Consider this: ✸ Wheels vs legs change where it can go ✸ Payload on paper ≠ payload in motion ✸ Runtime changes with load and terrain ✸ Top speed rarely means useful speed" / X](https://x.com/VaderResearch/status/2095226126033383928) Post Log in Sign up Post mete on X: "Buying a quadruped? Consider this: ✸ Wheels vs legs change where it can go ✸ Payload on paper ≠ payload in motion ✸ Runtime changes with load and terrain ✸ Top spe… - [@NVIDIARobotics on X](https://x.com/NVIDIARobotics/status/2095179991650644018) 🤖 A robot policy can be compact enough to run onboard and fast enough to keep the arm moving continuously. Post-trained NVIDIA Cosmos 3 Edge on NVIDIA Jetson AGX Thor T5000 generates roughly 2.13 sec… - [LimX Dynamics on X: "Together with ZINOVA's Tool Intelligence, TRON 2 takes on increasingly complex construction workflows. @ZiNovaLabs builds on TRON 2 to explore an innovative robotic configuration for construction, demonstrating key tasks in a scaled-down tilt-up construction workflow, includi… / X](https://x.com/chris_j_paxton/status/2095114381818253768) Post Log in Sign up Post LimX Dynamics on X: "Together with ZINOVA's Tool Intelligence, TRON 2 takes on increasingly complex construction workflows. @ZiNovaLabs builds on TRON 2 to explore an innovati… - [Noble Machines Accelerates Humanoid Robot Development 3X | NVIDIA](https://x.com/NVIDIARobotics/status/2094923298589024698) Manufacturing \| Robotics Noble Machines Accelerates General Purpose Industrial Robot Development With NVIDIA Isaac GR00T Learn More Objective Noble Machines is advancing industrial robot development … - [Seeed reBot Arm B601-RS: Physical AI & VLA Model Course | NVIDIA DLI Series](https://x.com/NVIDIARobotics/status/2094850247184826558) Warehouse China Warehouse US Warehouse Germany Warehouse The store will not work correctly in the case when cookies are disabled. US Warehouse: Enjoy FREE UNIUNI shipping on orders under $50! (Exclude… - [The Imitator Game — Benchmarking Robot Imitative Ability Beyond Action Prediction](https://x.com/XunzheZhou/status/2094530835215258044) Intent imitation, L0-L320,000+ paired episodesHuman evaluation Arena The Imitator Game BenchmarkingRobotImitativeAbilityBeyondActionPrediction Xunzhe Zhou 1,2,\,† · Yiyang Cai 2,3,\ · Fengyi Wang 2,3,… - [How to Train a Cross-Embodiment Robot Navigation Policy with AI Agents | NVIDIA Technical Blog](https://x.com/NVIDIARobotics/status/2093381258243895657) Technical Blog Subscribe Related Resources Robotics English한국어中文 How to Train a Cross-Embodiment Robot Navigation Policy with AI Agents Aug 26, 2026 By Yan Chang, Mihir Acharya, Wei Liu, Katie Washaba… - [Skild AI on X: "We found that to match the accuracy that S1 can achieve with just one example of prompting, current VLA models would need to be post-trained with 50-100 hours of data collection followed by fine-tuning!" / X](https://x.com/SkildAI/status/2092300859317330378) Post Log in Sign up Post Skild AI on X: "We found that to match the accuracy that S1 can achieve with just one example of prompting, current VLA models would need to be post-trained with 50-100 hours … - [Skild AI on X: "Introducing S1, our new foundation model that learns from one example. It can be taught 10-minute long tasks that it has never seen before, from one video prompt without any fine-tuning. Watch S1 operate in real-time via in-context learning:" / X](https://x.com/SkildAI/status/2092300842900865389) Post Log in Sign up Post Skild AI on X: "Introducing S1, our new foundation model that learns from one example. It can be taught 10-minute long tasks that it has never seen before, from one video prom… - [RoboPapers on X: "Instead of choosing between training a world model and training a language conditioned robot policy, why not do both? LDA-1B s a new foundation model that is trained on 30,000 hours of human and robot interaction data. Part of the secret is that LDA-1B jointly learns forward … / X](https://x.com/chris_j_paxton/status/2091877313872834800) Post Log in Sign up Post RoboPapers on X: "Instead of choosing between training a world model and training a language conditioned robot policy, why not do both? LDA-1B s a new foundation model that is… - [@axisrobotics on X](https://x.com/axisrobotics/status/2091797746114244876) Axis Weekly Last week, we closed the remaining replay and runtime gaps between the browser, policy server, and physics stack — then scaled the coverage of our task generation engine (TaskGen). An arti… - [Post-Train NVIDIA Cosmos 3 Edge for On-Device Robot Control | NVIDIA Technical Blog](https://x.com/NVIDIARobotics/status/2095179993974260168) Technical Blog Subscribe Related Resources Robotics Post-Train NVIDIA Cosmos 3 Edge for On-Device Robot Control Aug 19, 2026 By Saeed Babamohamadi +12 Like Discuss (0) - L - T - F - R - E AI-Generated… - [5 Physical AI infrastructure platforms shaping robotics in 2026 - The Robot Report](https://www.therobotreport.com/5-physical-ai-infrastructure-platforms-shaping-robotics-in-2026) Facebook X LinkedIn Reddit Pinterest Share From accelerated computing and simulation to data operations, open-source tooling, validation engineering, and continuous learning, these five platforms repr… - [Gemini Robotics 2 brings whole body intelligence to robots — Google DeepMind](https://x.com/SkillupAi/status/2093896499465945235) Skip to main content July 30, 2026 Models Gemini Robotics 2 brings whole body intelligence to robots Carolina Parada Share Your browser does not support the video tag. From feet to fingertips — we are… - [NVIDIA/Isaac-GR00T](https://github.com/NVIDIA/Isaac-GR00T)
NVIDIA Isaac GR00T N1.7 Header