Skild AI (@SkildAI) header
Skild AI (@SkildAI) profile picture

Building general purpose robotic intelligence.

USA
Joined June 2023
Photo by Skild AI (@SkildAI)Photo by Skild AI (@SkildAI)Photo by Skild AI (@SkildAI)Photo by Skild AI (@SkildAI)Photo by Skild AI (@SkildAI)Photo by Skild AI (@SkildAI)Photo by Skild AI (@SkildAI)Photo by Skild AI (@SkildAI)Photo by Skild AI (@SkildAI)Photo by Skild AI (@SkildAI)
Thanks! Been busy scaling deployments among other things. And now slowly sharing the ingredients behind our recipe. Hope the open source community takes it to next level. :)
there was lots of speculation at CoRL about what Skild was actually doing. I don’t think we should be so surprised 😉
2
3
54
4,610
Most viral robotics demos look identical to videos from 6 years ago. Yesterday at South Park Commons, Abhinav Gupta (Co-Founder & President, @SkildAI) broke down Physical AI without the hype. 5 takeaways:
2
1
2
42
7,869
Steel getting tougher day by day 🦾
Steel 0.2 is out: • 81 simulation environment examples, interactive docs, and migration instructions for your coding agent. • Magnetic contacts that latch, hold, and release. • Closed-loop joints for grippers and parallel robots.
1
2
27
2,052
What would it take to learn general manipulation entirely in sim? Accuracy, scalability, and diversity. Steel is our new physics solver, and it's (1) accurate. Each week we close the sim2real gap even more, this week on magnetic contacts and closed-loop joints. It's (2) scalable. Heterogeneous environments are painful to run at scale. So we worked with @NVIDIARobotics to take full advantage of the GPU, reaching speeds >10x faster than PhysX. Now, coding agents have unlocked environment creation at scale. We built Steel for them. Today we're releasing agent-first docs and examples to make it even easier. So (3) — it's diverse. There's never been a better time for sim.
Steel 0.2 is out: • 81 simulation environment examples, interactive docs, and migration instructions for your coding agent. • Magnetic contacts that latch, hold, and release. • Closed-loop joints for grippers and parallel robots.
14
16
173
9,470
Agents are very good at learning from examples, so we added more of them to our website Closed loop linkages, solver coupling, hydroelastic contacts, mimic joints, and so on!
Steel 0.2 is out: • 81 simulation environment examples, interactive docs, and migration instructions for your coding agent. • Magnetic contacts that latch, hold, and release. • Closed-loop joints for grippers and parallel robots.
6
54
3,811
Skild AI retweeted
We're hosting @SkildAI co-founder @gupta_abhinav_ at SPC in a few weeks. Bring your questions about the future of robotics. Request a spot below.
Today we’re open-sourcing Steel, our internal physics solver built with @NVIDIA. It runs simulations up to 13× faster than PhysX while using ~3× less GPU memory. Steel unlocks scaleable simulation for manipulation, for everyone.
1
3
7
2,203
A quick Fable 5.2 + Steel 0.1 test “Make Wuji hand climb a spiral staircase by walking on its fingers as legs” I feel like I am not ambitious enough 😁 github.com/skild-ai/steel-we…
2
5
62
4,719
We just built a rigid body solver that scales! Simulation is a crucial component of robotic infrastructure, and I believe we should build it in open Check out the blog, give it a try in Newton and send PRs my way, it’s time to accelerate🏎️💨
Today we’re open-sourcing Steel, our internal physics solver built with @NVIDIA. It runs simulations up to 13× faster than PhysX while using ~3× less GPU memory. Steel unlocks scaleable simulation for manipulation, for everyone.
5
10
1
124
7,801
Scalable simulation for robotics is here. Coding agents can now generate environments at scale. The next bottleneck? Simulating them at scale. Introducing Steel: a scalable physics solver for heterogeneous environments. Our first open-source release at @SkildAI, built in partnership with @NVIDIA. Our ambition is to rally the industry around closing the sim2real gap. Contributions welcome from you, your organizations, and your agents! ❤️
Today we’re open-sourcing Steel, our internal physics solver built with @NVIDIA. It runs simulations up to 13× faster than PhysX while using ~3× less GPU memory. Steel unlocks scaleable simulation for manipulation, for everyone.
28
35
9
349
21,861
Skild AI retweeted
Really excited to finally introduce Steel, now fully open source in Newton! Simulation infrastructure will be crucial to unlocking the scale of diverse experience needed for truly general robot manipulation. Building it in the open means anyone can try applying it to their own scenarios and as a collective everyone can push it forward. Come build with us! We’re improving it every day, have already adopted it internally, and we’re pushing for Steel to become the go-to solver for large-scale RL. github.com/skild-ai/steel
Today we’re open-sourcing Steel, our internal physics solver built with @NVIDIA. It runs simulations up to 13× faster than PhysX while using ~3× less GPU memory. Steel unlocks scaleable simulation for manipulation, for everyone.
5
6
1
56
2,541
Skild AI retweeted
Enjoyed hosting @deepakpathak , Co-founder & CEO of our portfolio company @SkildAI, at @TechCrunch Disrupt yesterday. One of my favorite moments was Deepak reflecting on what he’s had to unlearn as a founder. He once believed you could build a general-purpose robotics platform and leave deployment to others. The reality is far messier. Deployment is an integral part of the research loop, not something you can outsource while the technology is still maturing. Equally telling was his observation that customers care most about how quickly a robot can be deployed and deliver value. That makes in-context learning a critical piece of scaling general-purpose robotics. Skild recently crossed $100M in ARR, and they’re just getting started. #PhysicalAI #Robotics
2
3
2
19
16,130
Thanks, @Nasdaq, for celebrating @SkildAI's revenue milestone. Skild robots handle logistics, assemble GPUs, package food, and much more. Demos and free POCs are easy; getting someone to pay for it is hard!
Thanks to @Nasdaq for celebrating our 100M ARR milestone in Times Square.
6
7
3
106
7,741
Dynamic, interactive whole body control with a human. Humanoid robots are starting to use their bodies to interact with things much like the rest of us
We trained a robot to play football. How? Self-play for 140 years in a virtual World Cup. Meet the #Messinator:
2
3
28
4,741
AlphaGo, but for humanoid soccer. Skild AI brings self-play to humanoids. Their flagship robotics foundation model, S1, can now be post-trained with no human demonstrations by competing against itself in simulation. The testbed is soccer, which demands balance, agility, ball control, reactivity, and strategy. Given a single objective (score goals), S1 played 140+ years in NVIDIA Isaac Sim and went from falling over to dribbling past defenders, shielding, tackling, shooting, and getting back up. None of these behaviors were hand-rewarded; they emerged because they helped it score. Its opponents (past versions of itself) improved alongside it, creating an automatic curriculum. With 4 agents, passing and coordination started to emerge. Skild says the approach extends to social navigation, collaborative manipulation, and any multi-agent task with a simulator and an objective.
21
41
10
332
23,579
Skild AI retweeted
Think self-evolution is really a key towards professional-level sports intelligence…. Really great results!
We trained a robot to play football. How? Self-play for 140 years in a virtual World Cup. Meet the #Messinator:
2
11
2,352
Humanoid football had a big year. For comparison, this is what the world finals of RoboCup 2026 looked like just last month in South Korea. Should we compete next year? ;) nitter.cf/SkildAI/status/2102807…
We trained a robot to play football. How? Self-play for 140 years in a virtual World Cup. Meet the #Messinator:
16
77
3
334
73,822
Found the next G.O.A.T. It just needs charging. 🐐🔋
🤖 Made with AI
We trained a robot to play football. How? Self-play for 140 years in a virtual World Cup. Meet the #Messinator:
5
13
1,788
140 years of soccer practice in simulation. ⚽ Skild AI’s robot policy learned to dribble, shield, and tackle through self-play in NVIDIA Isaac Sim before being transferred to a real humanoid robot. 👇
We trained a robot to play football. How? Self-play for 140 years in a virtual World Cup. Meet the #Messinator:
29
61
6
574
42,042
Skild AI retweeted
This is one of the most fun projects I've worked on. Playing against the robot was surreal. It's surprisingly fast! We tried many approaches, but the simplest worked best at scale: self-play on top of a skill-conditioned low-level policy.
We trained a robot to play football. How? Self-play for 140 years in a virtual World Cup. Meet the #Messinator:
4
7
108
5,466
Skild AI retweeted
Skild's S1 stands out because it showed more pretraining data made the model substantially better at performing unseen tasks from a single video prompt, without fine-tuning. A stronger foundation model made each new demonstration more useful. Now, Skild is showing a complementary approach: generating experience that humans never have to demonstrate. Skild post-trained a soccer policy through 140 years of simulated self-play against earlier versions of itself. No human demonstrations during this post-training phase. Behaviors like shielding the ball and tackling emerged without separate rewards engineered for each skill. The policy then transferred to a physical humanoid. AlphaGo was such an important moment for AI because human expertise could be a starting point rather than the ceiling. A system could discover strategies nobody had explicitly taught it. Self-play creates a powerful learning loop. As the model improves, it becomes a tougher opponent for its next version. Progress generates the next challenge. More compute can produce new experience, rather than just more training on a fixed dataset. What’s interesting here is applying it on top of a general robotics foundation model, with behaviors transferring into the physical world. Together, Skild’s results suggest two complementary ways to scale: learn more effectively from the experience humans provide, then improve through experience the model generates itself. Soccer has clear objectives and measurable outcomes. Extending this to everyday work, and showing that the gains generalize beyond a single task, is the next test. Skild is showing how a capable foundation model could help change that constraint of human capability: human demonstrations get it started, without having to supply every subsequent improvement.
We trained a robot to play football. How? Self-play for 140 years in a virtual World Cup. Meet the #Messinator:
3
1
29
4,271