We’re live at @NVIDIAGTC. 🚀
All week at Booth 1406 right by the entrance to Hall 2, the energy around @LightwheelAI has been incredible.
If you’re at @NVIDIAGTC, come say hi.
English
Lightwheel
122 posts
Lightwheel
@LightwheelAI
Simulation Infrastructure for Physical AI
Santa Clara, U.S.A. Beigetreten Şubat 2024
49 Folgt1.2K Follower
@NVIDIAGTC RoboFinals Is Being Adopted by Leading Robotics Teams
As robotics foundation models scale, evaluation is becoming the bottleneck of Physical AI. Training is advancing rapidly, but the infrastructure to systematically measure robot capability is still limited.
That’s where RoboFinals comes in.
RoboFinals is @LightwheelAI evaluation platform for Physical AI, built on the top of NVIDIA Isaac Lab - Arena with @NVIDIARobotics, designed to benchmark robot policies across tasks, robots, and environments at scale.
A growing number of robotics teams are already using RoboFinals, including:
• @Alibaba_Qwen – evaluating embodied foundation models
• @FourierRobots – testing humanoid robot policies
• @roboforce_ai – stress-testing industrial robotics systems
• @peritas_ai – validating safety-critical medical robotics
At the core is RoboFinals-100, an industrial benchmark spanning household, factory, and retail environments with interactions across rigid, articulated, and deformable objects.
To scale evaluation workflows further, we built AutoDataGen, an automated synthetic data generation pipeline introduced at NVIDIA GTC. AutoDataGen uses LLM-driven task decomposition to automatically generate action data for benchmark scenarios.
Under the hood, RoboFinals runs on a scalable infrastructure stack:
• SimReady physics-grounded simulation environments
• RoboFinals-100 benchmark tasks
• NVIDIA Isaac Lab-Arena modular evaluation framework
• NVIDIA OSMO orchestration for distributed workloads
Together with large-scale GPU clusters, RoboFinals can run thousands of evaluation episodes in parallel.
Our ambition is clear:
RoboFinals aims to become the ImageNet of robotics evaluation.
Learn more: lightwheel.ai/media/robofina…
#PhysicalAI #NVIDIAGTC #NVIDIARobotics #Robotics #EmbodiedAI #RobotLearning #Simulation #OpenUSD
@jonstephens85 , @lukas_m_ziegler , @IlirAliu_
English
Appreciate @HumanoidRTech for featuring @LightwheelAI at @NVIDIAGTC!
That is the direction @LightwheelAI is leading through our work across the Newton ecosystem, from SimReady standards and solver advancement to real-to-sim workflows and physically measured reference assets.
Humanoid Robotics Technology@HumanoidRTech
.@LightwheelAI has joined the Newton Technical Steering Committee to help lead the future of physically grounded simulation. As part of the Newton initiative under The Linux Foundation, alongside @nvidia , @GoogleDeepMind, Disney Research, and @ToyotaResearch, Lightwheel is leading core efforts shaping the next generation of simulation infrastructure for Physical AI. As simulation becomes foundational to real industrial and embodied AI workflows, the challenge is no longer just scale. It is physical accuracy, reuse, and deployment readiness. These are the infrastructure problems that will determine whether simulation can truly support robot training, evaluation, and real world deployment, and they are exactly where Lightwheel is helping drive the ecosystem forward. Lightwheel’s work across Newton includes: • Leading SimReady schema evolution and technical direction • Driving solver advancement and real to sim calibration workflows • Building physically grounded tooling for real to sim data generation • Delivering physically measured, solver validated reference assets and environments This is the infrastructure layer needed to move simulation closer to real world physics and support the next wave of Physical AI. Learn more here: developer.nvidia.com/blog/newton-ad… If you’re at NVIDIA GTC today or tomorrow, stop by Booth 1406 to learn more about what Lightwheel is building. #PhysicalAI #NVIDIAGTC #Robotics #EmbodiedAI #Simulation #HumanoidRobotics @NVIDIAAI
English
Day 3 at @NVIDIAGTC is off to a packed start at @LightwheelAI Booth 1406 🔥
The energy at the @LightwheelAI booth is already incredible, with robotics teams, researchers, founders, and partners stopping by for our Physical Measurement Factory, live demos, and simulation evaluation platform.
Excited to also welcome @RavitJain from The Ravit Show to the booth for an interview!
More than booth traffic, this feels like a signal of where Physical AI is heading: toward simulation that is physically grounded, measurable, and ready for real world deployment.
If you’re at GTC, come find us.
#NVIDIAGTC #PhysicalAI #Robotics #EmbodiedAI
English
Live today at @NVIDIAGTC!
Join us at Booth 1406 to see how @LightwheelAI is building the infrastructure for Physical AI.
Featured at the booth today:
◦ Physical Measurement Factory
See how real world physics is measured and translated into simulation ready assets.
◦ Live Demos
Explore physically grounded simulation workflows for robot training and deployment.
◦ RoboFinals Evaluation Platform
See how frontier robotics teams are evaluating model performance at scale.
From physically accurate asset to industrial-grade evaluation, Lightwheel is defining the stack for physically trustworthy robotics simulation.
Stop by Booth 1406.
#GTC26 #NVIDIAGTC #PhysicalAI #Robotics #EmbodiedAI #Simulation
English
@NVIDIAGTC: Inside @LightwheelAI's Industry-Leading SimReady Architecture
At NVIDIA GTC, we’re excited to pull back the curtain on the methodology we use to build the highest quality simulation assets at scale: Measure → Solve → Generate.
In our new blog, we reveal how this architecture transforms real world physics into physically grounded simulation assets at scale for robot learning, data collection, and evaluation.
Read more: lightwheel.ai/media/simready
We’ve also brought our Physical Measurement Factory to the @NVIDIAGTC show floor, so builders can see firsthand how real world behavior is measured and translated into simulation ready assets.
For Physical AI, assets cannot just look real. They need to behave real.
Come by Booth 1406 to see the Physical Measurement Factory in action.
#NVIDIAGTC #PhysicalAI #Robotics #Simulation
English
@LightwheelAI at @NVIDIAGTC : At the Center of the Robotics Spotlight ✨
At NVIDIA GTC, @jonstephens85 , Chief Evangelist at @LightwheelAI, was featured in a standout robotics session alongside leading voices from across the industry.
Physical AI is moving from research milestones to real-world deployment, and Lightwheel is at the center of that shift. At NVIDIA GTC, Jonathan brought that momentum to one of the industry’s most visible stages.
Great to share the stage with @akhil_docca of @nvidia , @mack2it of @workr_official , Rajesh Radhakrishnan of @ServeRobotics , @kevinmpeterson1 of @BedrockRobotics, and @Yao__Lu of @physical_int.
#NVIDIAGTC #PhysicalAI #Robotics #EmbodiedAI #Leadership #Ecosystem
English
@LightwheelAI at @NVIDIAGTC: Teleoperation as Learning Infrastructure for Physical AI
At GTC, Jay Yang, Chief Architect at @LightwheelAI, shared our perspective on teleoperation for Physical AI:
Teleoperation is not the end goal. Scalable learning is.
Raw teleoperation data is not training-ready by default. To support robot learning at scale, teleoperation must evolve into infrastructure with unified action semantics, reproducible replay, consistent sim-and-real interfaces, and distributed data pipelines.
We are building this stack end-to-end, from teleoperation and retargeting to evaluation infrastructure, benchmark orchestration, and cloud-scale data generation.
Thanks to Maarten Witteveen of @ManusMeta and Jiwen 'Steve' Cai of @nvidia for the great discussion.
#NVIDIAGTC #PhysicalAI #Robotics #EmbodiedAI #Teleoperation #Sim2Real #RobotLearning
English
@LightwheelAI at @NVIDIAGTC: Accelerate Physical AI with Simulation-Centric Data Engine
At GTC, Myles Liu, Director of Business Operations at @LightwheelAI, shared how Physical AI is moving beyond research demos and why scaling it requires a new kind of infrastructure.
At Lightwheel, we believe Physical AI is built on three core pillars: World, Behavior, and Eval.
• World defines physically grounded simulation environments.
• Behavior provides scalable synthetic data and egocentric data for learning.
• Eval measures whether models actually improve in ways that matter for real-world deployment.
That is the infrastructure layer Lightwheel is building.
From production-grade SimReady assets and scalable scenes, to large-scale synthetic and egocentric data generation, to RoboFinals for industrial-grade policy evaluation with Isaac Lab-Arena, Lightwheel is building the simulation-centered data engine for scalable Physical AI.
Great to share this vision at GTC and excited to keep pushing the industry forward.
#NVIDIAGTC #PhysicalAI #Robotics #EmbodiedAI #Simulation #SyntheticData #Evaluation
English
@LightwheelAI at @NVIDIAGTC: Solving the Three Computer Problem for Physical AI
At GTC, @HeLIAN_AEO, VP of Partnerships and Strategy at Lightwheel, joined @evanhelda of @nebiusai , Bill Vass of @BoozAllen, @LerrelPinto of Assured Robot Intelligence (ARI), and Jon Quick of Launchpad to discuss the “Three Computer” problem and what it takes to build real infrastructure for Physical AI.
At Lightwheel, we believe Physical AI will not scale through fragmented tools. It will scale through a simulation-centered infrastructure where assets, data, and evaluation work as one system.
That is where Lightwheel leads.
From physically grounded SimReady assets, to measurement-driven simulation workflows, to evaluation systems that turn failure into learning signal, we are building the infrastructure layer for scalable Physical AI.
Great discussion at GTC, and excited to keep pushing the ecosystem forward with leaders across cloud, robotics, and simulation.
#NVIDIAGTC #PhysicalAI #Robotics #EmbodiedAI #Simulation #Infrastructure
English
Want to see how @LightwheelAI is building the infrastructure layer for Physical AI?
In this booth walkthrough, our Chief Evangelist @jonstephens85 takes you inside the full Lightwheel stack at @NVIDIAGTC, from our ambitious vision for Physical AI, to our physically grounded measurement factory, live teleops demo with JoyLo, and industrial-grade simulation evaluation platform.
Want to learn more? Come see it in person at Booth 1406!
#NVIDIAGTC #PhysicalAI #Robotics #EmbodiedAI #Simulation
English
Lightwheel retweetet
Yesterday the keynote, today workshops! 👨🏼🏫
The awesome thing about the GTC is that it is not just about the tradeshow, or about the keynote.
They have so many awesome sessions and hands-on workshops to attend.
Today I stopped by @LightwheelAI hands-on lab, where we were building simulation-ready robotics workflows.
At their lab session, we went through the full pipeline of preparing simulation-ready assets for embodied AI, from raw data to deployed robot behavior.
The focus is very practical as it shows how to go from a static 3D scene to something a robot can actually learn from.
The workflow covered generating high-fidelity synthetic data using OpenUSD configuring physics properties (mass, friction, collision) with PhysX and Newton adding semantic metadata and manipulation affordances (so robots understand how to interact with objects).
The final step was validation, so deploying trained robot agents inside NVIDIA Isaac Sim to test whether everything behaves correctly.
It’s a good example of where robotics is heading, not just better models, but better infrastructure for training and validation.
Learning never stops! 🫡
~~
♻️ Join the weekly robotics newsletter, and never miss any news → ziegler.substack.com
English
Thanks @lukas_m_ziegler for co-hosting such an incredible night with us.
Great night yesterday. More to come!
Lukas Ziegler@lukas_m_ziegler
Santa Clara was fire! 🔥 Together with @LightwheelAI we hosted a party for the physical AI community. Once Steve Xie (@bgxc) reached out to make the coolest robotics event during GTC, I was like 'hold my actuator, we got it'! I knew there would be tons of people into robotics, simulation, and embodied AI at GTC, but I had no idea there would be so many of you! At the peak, there were over 350 of us! 🤯 We had dancing humanoids, fighting bots, lots of them! We even had a Cybertruck with a robot arm attached to it (I'm not kidding). Crazy energy, awesome people from researchers, to founders and operators to investors, A lot of robo-friends in one place. 🫶🏼 We even had a humanoid robot featuring at the DJ booth! That's everyone for coming! It was awesome to co-host such an amazing event! More parties to follow throughout the 2026, so stay tuned! ~~ ♻️ Join the weekly robotics newsletter, and never miss any news → ziegler.substack.com
English
Real2Sim for Industrial Robotics with real physics is live soon at @NVIDIAGTC
At 2:00 PM, @jonstephens85 from @LightwheelAI will lead a live discussion with @taoyu618 from Analog Devices on:
Bringing Physical Intelligence into Scalable Robotics Simulation: Real2Sim Validation for Industrial Robotics
At Lightwheel, we believe the next step for robotics simulation is not just scale, but physical validation.
In this talk, we’ll share a joint effort with ADI around sensor integrated simulation, physical measurement workflows, and evaluation driven infrastructure for industrial robotics and Physical AI. We’ll look at how real world force measurements, contact behavior, and deformable interactions can be used to evaluate simulation fidelity and identify where simulation matches physical reality and where it still falls short.
For teams building high precision industrial manipulation systems, this is a foundational problem for training, evaluation, and deployment.
Tuesday, March 17
2:00 PM
Lightwheel Booth 1406
Come by and join us.
#NVIDIAGTC #PhysicalAI #Robotics #IndustrialRobotics #Simulation #Real2Sim #EmbodiedAI
English
The @LightwheelAI booth is packed at @NVIDIAGTC 🔥
Teams are stopping by to see our physical measurement factory, live demos, and simulation evaluation platform in action. The signal is clear: the next frontier in Physical AI is not just building models, but building the infrastructure to measure, validate, and evaluate them against the real world.
Great conversations all day at the booth. If you’re at GTC, come find us at Booth 1406!
#NVIDIAGTC #PhysicalAI #Robotics #Simulation #EmbodiedAI #Sim2Real
English
Such a full room and strong engagement at @LightwheelAI training lab at @NVIDIAGTC 🔥🔥
High-quality robot learning starts with high-quality assets, and it’s clear this is becoming a central focus across the Physical AI ecosystem.
At the session, we broke down how robot-ready assets are built for simulation-driven training and real-world transfer, from physically measured asset development and solver validation to real-to-sim calibration workflows.
We also walked through how to:
• prepare robot-ready assets for training
• configure physical properties such as mass, friction, and collision geometry
• add semantic and affordance metadata for manipulation tasks
• validate assets in NVIDIA Isaac Sim with trained robot policies
• understand how simulation fidelity impacts policy behavior and sim-to-real transfer
Strong interest, thoughtful discussion, and clear momentum around simulation fidelity and asset quality as foundational layers of Physical AI.
@NVIDIARobotics #NVIDIAGTC #PhysicalAI #Robotics #Simulation #IsaacSim #Omniverse
English
First night at @NVIDIAGTC — the Lightwheel Physical AI Party.
350+ people showed up.
The room was full. The vibe was 🔥
Robotics, AI, simulation, infrastructure — everyone in one place.
A lot of new connections. A lot of great conversations.
And honestly, a lot of fun.
This is what the Physical AI community looks like.
And we’re just getting started.
#PhysicalAI #Robotics #EmbodiedAI #HumanoidRobots #RobotCommunity
#NVIDIAGTC #AICommunity
English
Proud to see Lightwheel featured in Jensen Huang’s NVIDIA GTC 2026 keynote.
As Physical AI moves from research to real-world deployment, simulation is no longer just a tool, it is becoming core infrastructure.
At Lightwheel, we are building this infrastructure layer through:
• Physically measured, solver-calibrated SimReady assets
• Scalable generation pipelines
• Industrial-grade simulation and evaluation platforms
Together, these systems bring simulation closer to real-world physics, enabling robotics teams to train, evaluate, and deploy with greater confidence and scale.
It’s especially exciting to see how broadly simulation is shaping the robotics workflows highlighted in the keynote, and to be supporting many of the teams pushing the frontier forward.
The future of robotics will be built on physically grounded simulation.
Lightwheel is building that foundation.
English
@NVIDIAGTC Behind Lightwheel’s SimReady Architecture: Measure → Solve → Generate
As robotics models scale, the challenge is no longer just training, it is producing physically accurate simulation assets at scale.
At @LightwheelAI , we are building simulation infrastructure for Physical AI to solve this problem.
Our SimReady System introduces a unified architecture for generating physics-grounded simulation assets through a closed-loop pipeline:
Measure → Solve → Generate
• Measure
Capture real-world properties such as friction, stiffness, deformation, and contact dynamics using our Physical Measurement Factory.
• Solve
Translate measured parameters into reliable simulation behavior through advanced physics solvers.
• Generate
Produce OpenUSD-based assets and environments at scale for robot learning, teleoperation data collection, and evaluation.
Together, this architecture enables physically accurate simulation assets at scale, grounding robot learning in real-world physics.
We are also collaborating with @NVIDIARobotics to advance simulation physics using the Newton Physics Engine, and enabling SimReady assets to integrate seamlessly with NVIDIA Isaac Sim and Isaac Lab workflows.
Learn more about the SimReady Architecture: lnkd.in/gjjkNy59
learn more about OpenUSD in Physical AI simulation in NVIDIA’s GTC talk: lnkd.in/g__xxm9X
#NVIDIAGTC #NVIDIA #NVIDIARobotics
English
Lightwheel retweetet
.@LightwheelAI is developing simulation infrastructure designed to accelerate Physical AI and humanoid robotics, a focus the team will be highlighting at @NVIDIAGTC 2026.
At Booth 1406, Lightwheel will be showcasing its latest work on simulation infrastructure for Physical AI. This includes:
✔️ Physical measurement and real-to-sim calibration for physically grounded SimReady assets
✔️ Physics solver development and calibration for complex interactions
✔️ Large-scale synthetic data generation pipelines for robot learning and evaluation
✔️ A simulation-driven evaluation platform for robotics (RoboFinals)
These capabilities help enable high-fidelity virtual environments for large-scale training, testing, and evaluation of next-generation humanoid robots, integrating technologies such as @nvidia Isaac Sim, Isaac Lab, and the GR00T foundation model.
Alongside GTC, Lightwheel will also be hosting a Physical AI gathering on Monday night in Santa Clara, bringing together founders, engineers, researchers, and investors across the robotics and Physical AI ecosystem. The evening will feature live robotics showcases, interactive demos, and curated networking designed to spark meaningful conversations across the rapidly growing Physical AI community.
Learn more about Lightwheel and register your attendance for their Physical AI Night here: humanoidroboticstechnology.com/events/physica…
#Humanoid #Humanoids #HumanoidRobot #PhysicalAI
English