Data and Simulation Infrastructure for Physical AI
Santa Clara, U.S.A.
Joined February 2024
- Tweets217
- Following76
- Followers2.6K
- Likes145
Huge congrats to our partners at @dreamlabs_ai on the launch of Dream-1!
One model. Any two-armed robot. No retraining.
Excited to see this out in the world!
Today, Dream Labs is coming out of stealth with Dream-1, a robot foundation model that controls bimanual robots it has never seen, on tasks it was never shown, in places it has never been, from $3/hour.
Dream API is live for everyone, today. 🧵
Heading to #GTCBerlin? Come find us after keynote day.
@LightwheelAI and @NVIDIARobotics are hosting the RoboFinals Party. Keynote day ends, the real conversations start. The people who build the models, run the simulators, and write the checks, one room, open bar.
📅 Oct 21 | 6–11 PM
📍 Berlin Mitte, 15 min from Tempodrom and STATION-Berlin
🍸 Open bar, curated food and drinks
🎟️ Space is limited and attendance is by approval. Request your spot: luma.com/mctmd6xm
#GTCBerlin #PhysicalAI #Robotics #RoboFinals #NVIDIA
Congratulations to the @naturewill team on the release of Natura-1!
Natura-1 is a 50B-parameter MoE foundation model for the physical world, pretrained on one million hours of egocentric human experience.
At Lightwheel, we’re proud to support Natura-1 through EgoSuite, providing high-quality human data at scale. Our data spans real-world tasks in homes, retail stores, hospitals, and factories, with precise bimanual hand-pose annotations and footage from wrist-mounted cameras.
Real-world human experience, captured in detail and delivered at scale.
#NatureWill #Natura1 #PhysicalAI #EgoSuite #HumanData
IROS 2026 is a wrap!
After a week of conversations, demos, and discussions across the robotics community, three trends stood out to us.
Here are our three main takeaways from the conference:
1. Dexterous manipulation and AI is maturing. We saw many different hand embodiments and tactile sensing displays at various booths. We saw this trend playing out earlier this year and have partnered with Analog Devices to help pave the future of dexterous AI.
2. Deployment is everything! We saw a lot of interesting robots, but the main question is: who has deployed them? This is why we had Vega from Dexmate making hats in our booth. We partnered with Dexmate and Paritas AI to deploy Vega in hospitals, solving labor gaps in surgical tool kitting.
3. Simulation is back! Well, it never went away. But interest was high around our RoboFinals eval solution and Newton-native simulation assets. As Physical AI scales, so will the need for quality evals.
Didn't get a chance to stop by our booth or attend our party?
Visit us at CoRL next month!
#PhysicalAI #IROS2026
We’d love your vote! 🗳️
Lightwheel has been named a finalist for Best Use of AI in the 2026 Humanoid Robotics Industry Awards.
We’re proud to be recognized alongside so many companies pushing humanoid robotics forward.
If you’ve followed our work across Physical AI, simulation, and robot learning, we’d appreciate your support.
👉 Vote for Lightwheel here:
humanoidroboticstechnology.c…
Voting is open through October 15.
#Robotics #PhysicalAI
It has been a busy week at #IROS2026! Today is your last day to stop by our booth to learn about our open sourced 100K hours of egocentric data, our next generation of benchmarking in sim, and even get a custom @LightwheelAI hat made by @DexmateAI's robot!
#Robotics #PhysicalAI
Lightwheel is heading to #IROS2026. Booth 714, Pittsburgh, Sep 27 to Oct 1.
🔹 Learn: EgoSuite-Open100K, 100K hours data of humans doing diverse and real tasks
🔹 Train: SimFoundry, simulation worlds where robots practice
🔹 Evaluate: RoboFinals, a standard exam for robot policies
🔹 Deploy: RoboStack, from sim to real robot deployment
After hours, join our joint party with @ADI_News and @DexmateAI.
Sep 29, 6 to 10PM, Pittsburgh. Bring your hardest sim-to-real question.
RSVP: lightwheel-iros.netlify.app/…
Lightwheel is hiring for 2027.
We build the data and simulation infrastructure for Physical AI. Come help us take it global.
Roles
→ Strategic Product Manager, Early Career (GTM)
→ Strategic Projects Manager (Global Operations)
Who can apply
→ Full-time: graduating Sept 2026 to Aug 2027, in CS, Robotics, Automation, AI, or related fields
→ Internship: current students
📍 SF Bay Area
🔗 lightwheel.ai/careers
#PhysicalAI
Sim to real. Summit to party. 🎉
The humanoid robotics world is in Seoul next week, and on Wednesday night it's all in one room. Builders, researchers, and investors, off the conference floor and talking sim2real. 🤖🍻
It caps our day at @HumanoidsSummit, after @bgxc's Opening Keynote and demos at Booth 16-21.
🗓 Sep 23, 5–10 PM
📍 Seoul (venue shared upon RSVP)
Spots are going fast. Apply now 👉 lightwheel-seoul.netlify.app…
Lightwheel is at @HumanoidsSummit Seoul, Sep 22–23 at COEX 🇰🇷
Sep 23, 9:00 AM: @bgxc delivers the Opening Keynote, "Building the Continuous Learning Infrastructure for Physical AI."
Booth 16-21: see the full loop, from human data to deployment. Our team is on-site all week
The best conversations at IROS never happen in the sessions. They happen later that night, plate in one hand, with the people who make this field worth being in.
@LightwheelAI x @ADI_News x @DexmateAI are co-hosting a party on September 29, 6-10PM, in Pittsburgh. Good food, a proper bar, four hours off the clock.
Bring your hardest sim-to-real question. We'll handle the rest!
#IROS2026 #Robotics #PhysicalAI #Dexmate #AnalogDevices
We started in simulation. Egocentric data is where that led us.
Our co-founder @bgxc explains on @RoboPapers why scaling sim means scaling real data first, and why egocentric data was the only source that could be standardized.
Watch the full episode with @micoolcho and @DJiafei here: youtu.be/ZywjONaTHRg?si=bAN5…
#Robotics #Real2Sim
Learn More:
Project page:
egosuite100k.lightwheel.ai
Blog post:
huggingface.co/blog/Lightwhe…
Data on HuggingFace:
huggingface.co/collections/L…
Why does a simulation company open-source 100,000 hours of egocentric data?
Because training data is scaling fast and evaluation isn't. @bgxc explains on RoboPapers: ego data is the most scalable way to train, simulation is the only scalable way to test. EgoSuite-Open100k is half of that; RoboFinals is the other half.
Thanks @micoolcho and @DJiafei for the conversation.
Robotics has a data problem, and egocentric data is a compelling way to solve it because egocentric data is (1) scalable — it’s cheap to collect — and (2) easily captures the breadth and diversity of real human tasks, while (3) being true to real physics. Lightwheel recently released a hundred thousand hours of high-quality egocentric video data, spread across a wide range of human tasks from cleaning and tidying to assembly and construction to retail tasks.
Here to tell us about this tremendous effort is @bgxc, CEO and co-founder of Lightwheel. The data is also available on HuggingFace right now.
Watch Episode 102 of RoboPapers, with @micoolcho and @DJiafei, to learn more!
Lightwheel retweeted
Full episode dropping soon!
Geeking out with @bgxc on EgoSuite-Open100K egosuite-oepn-100k-test.ligh… @LightwheelAI
Co-hosted by @micoolcho @DJiafei
We were on the ground this week at LEAP x DeepFest AI 2026.
Martin Elbs spoke on Building the Continuous Learning System for Physical AI, and how closing the loop between data, simulation, evaluation, deployment, and real-world feedback can accelerate how robots learn.
Louis Lian joined the panel How Generative AI is Redefining Robotics at Scale, discussing how advances in generative AI are changing how we train, evaluate, and ultimately deploy robots at scale.
And the Lightwheel team is at the booth connecting with the Physical AI community. Great conversations and a lot of momentum around the future of robotics.
If you missed us, we'll at @HumanoidsSummit Seoul in a few weeks!
#LEAPxDeepFest2026 #PhysicalAI #Robotics
