Simulate, train, deploy, and fund robots from a single stack.
- Tweets 697
- Following 13
- Followers 8,981
- Likes 2,116
The launchpad of @machinedotfun isn't only for robots.
We're looking for hardware teams across a range of industries: energy, autonomous systems, gaming, mobility, smart home, wearables, industrial tools, and more.
If you're building something physical, we want to see it.
2
10
102
7,040
We are now on @fomo.
Follow along as we build the infra powering the next generation of robots.
10
11
111
7,420
Building is hard. Manufacturing and shipping is harder.
The Foundry by @machinedotfun handles the second part.
3
7
66
3,128
Martin from @peaq remembers the first time he met us.
A robotics team that launched on @pumpfun, with no VC backing.
That's where @machinedotfun started.
4
14
99
4,308
SIMARENA IS COMING TO MUNICH 🏰
We're partnering with robo.innovate for their 2026 hackathon, alongside @nvidia, @Universal_Robot, @NEURARobotics, @peaq, Doosan Robotics, @Teradyneinc, @Microsoft, and more.
SimArena will be used across the major challenges each partner is putting in front of the young builders.
12
13
100
9,916
Finally have a reason to own the iPhone Duo.
Simulate on one screen, launch on the other.
6
7
72
2,931
Robots used to be trained on real hardware, so every mistake broke something. Now that part happens in a browser tab.
SimArena is the first step, @machinedotfun is the next, and between them a machine becomes a business instead of a hobby.
4
11
73
7,425
A tracker can follow a ball perfectly, then meet it as a stranger the second a cup covers it.
Object permanence isn't free. Classic trackers bolt it on. World models and VLAs learn it.
5
49
2,052
Full breakdown of the Microduck by @pollenrobotics.
Bipedal and self-balancing on fifteen smart servos, with a single Rockchip RK3566 board handling the onboard compute, control loop, and AI inference.
state of the art robotics.
1
8
50
2,289
Microduck is now available in SimArena.
Play with it from your browser, make it pull some skills, then record a clip and drop it straight on your timeline. These are the default skills, more coming soon.
app.simarena.ai/play/microdu…
4
6
50
1,936
Teaching a robot to walk means building the world it learns in first. You spent two years as a toddler falling over on every floor in your house. A robot doesn't get two years, so it gets a simulator, and it needs every floor you can think to put in there.
1
4
39
1,382
"Large language models learn from the internet, but robots don’t have an internet-scale dataset of physical experience. What’s the equivalent for robotics, and how do we get it?"
Here’s what @chelseabfinn, co-founder of @physical_int, has to say.
3
4
31
1,577
Robotics simulation isn't going to settle on one engine. Different engines are good at different parts of the same robot, and a single robot passes through several of them before it works.
Picking a winner was never the point. What you actually need is one place to work across all of them. That's the layer SimArena is going after.
2
9
40
2,839
Good night at RoboEve yesterday.
Room full of people building in robotics and others just getting into it, the mix we hoped for. A few even brought their own hardware.
We introduced Foundry to them in person. Thanks to everyone who came. @SuperteamAE @solana
6
7
62
3,092
We're hosting Robo Eve on Thursday, 23 July in Dubai.
Join us for a night exploring physical AI: the hidden layer beneath it, simulation, training data, and the people building it. All in one room.
If you're building in robotics or AI, or want to be, don't miss it.
RSVP: luma.com/mb4h6tqy
1
7
36
1,670
There are two ways to make a robot act, and the field hasn't picked one yet.
A VLA takes what the camera sees, what you tell it, and the robot's own state, and maps it straight to an action. A trained reflex, learned from huge robot datasets. π0.5 and SmolVLA work like this. It's fast and simple because almost nothing sits between seeing and moving. The weakness is that it only knows what it learned. Push it far enough outside its training and it can fail, because it carries no model of what happens next.
A world model goes the other way. It learns to predict how the world changes, then uses that learned simulator to test moves and imagine outcomes before the robot commits. Dreamer is built this way, and DayDreamer put it on real robots. The strength is foresight: it can reason over possible futures instead of repeating past actions. The cost is compute and its own mistakes, because every imagined future has to be predicted, and a bad prediction means a bad move.
So the trade is reflex against foresight. They aren't rivals though. World models are increasingly used to generate experience and training data for the action models. The likely future is hybrid, where the imagination teaches the reflex.
3
8
40
1,660
Once robots can plug into a model and start acting in the world, a second question shows up fast: who owns the machine, and how does it earn.
@MartinElKhouri, CBO of @peaq, has been building exactly that layer. He's joining CodecFlow as an advisor on business and strategy.
10
10
65
10,253
Show a robot a task and it'll copy you move for move.
Move one thing, and the copy falls apart. That's behavior cloning, and it's why simulation matters: you let it see the task a thousand different ways before the real world rearranges it.
7
31
1,657
Claude Fable 5 + Hermes one-shotted a documentary on how far robots have come.
One single prompt: researched the history, pulled the archive footage, sequenced it, scored it, cut it, and ran narration through Higgsfield MCP.
Robots were first judged on whether they could move, then whether they could work, then whether they could decide.
Next test is trust. And trust gets earned in simulation. SimArena.
3
8
46
3,375
CodecFlow × @MCGlive stream today.
We’ll discuss robotics, SimArena, and what’s coming next.
12:30 PM EDT · 16:30 UTC
26 May
Lock in.
1
4
38
2,693
SimArena is the on-ramp to CodecFlow's robotics stack.
Coming next: world-model scenes, AI agents, chat-to-build scenes, sensor sim, digital twins, URDF builder, headless API, and more.
1
2
35
2,219
