David Chen

@chetan_ I like how it threw the pillow into the corner.

3 thoughts. 1. The whole body manipulation of the basket is dope. It's easy to forget how much humans use this. 2. Glad they kept audio and 1x speed. The robot is still loud but way less stomping then a year ago. 3. Still a demo and still mostly pick and place

Come join us at our next Physical AI Meetup, where we'll be discussing the challenges of deploying robots in homes with @tonyzzhao from @sundayrobotics, @itsdanielho from @1x_tech, and @kaandogrusoz from @weaverobotics! 🤖🤖🤖 luma.com/physical-ai-me…

@fdotinc Artifact was incredible. Lots of robots, hardware, drones, hoverboards?!

@robotsailor Dope

introducing our new rover today: -new design -new electronics -extremely intelligent more soon.

@JohannesTscharn @Spectacles Very cool. Tell Claude to integrate LiveKit Agent for the voice interaction loop.

Hooked up Reachy to an LLM and made a couple of tools like detect_objects or look_at It can now essentially reason about the space it’s in by using @Spectacles AR TTS latency is still high and needs a better solution, Claude 4.6 is already working on it 😎

as promised, second part on learning world model learning is up github.com/pham-tuan-binh… the focus is on inverse dynamics this time, arguably the most important component of modern world models this comes from a very first principle question “how can we train a world model without action labels” again, we go through stuff like: > different conditioning strategies (mentioned in DiT) like adaptive layer norm to just simple addition > avoiding modal collapse through masking and more practical stuff when it comes to the reconstructive loss function of an inverse dynamics model vs a dynamics model p/s: i have to say, the progress on world models has been absolutely insane during the time i write this. for example, two weeks ago, nvidia releases cosmos policy which has no inverse dynamics model, no action head, only one single video world model to rule them all. then just this week, they dropped a bombshell called dreamzero to show zero shot behavior on world model based planning. on a practical note, you'll be surprised on how much inverse dynamics is NOT used in the field, but that is for another day.

@i2cjak Where can I find this?

holy shit they've done it

@AnselDias Have you tried stereo cam to VR? The depth sense actually helps a ton.

First time doing camera-only teleop. I was trying to see how well I can pilot it without looking directly at the scene. If I can't do it, neither can the AI, right? Currently I just have a 4k 60FPS monocamera on it making depth perception hard. Plan to add wrist cams soon.

@mattfreed Where’s it gonna be?

Culture shift in SF as more hardware folks come in If you work on hardware come grab a beer Thursday realhardwarepatriots.com

@Sethnuzum I'll def play with the free version to get a better sense of the features!

@chenosaurus It can be, but I was actually just using the free version with the watermark! Definitely try to build something better with Claude and let me know. Would love another option.

I projection mapped my living room last weekend. Crazy what you can do in 20 minutes with free software and a $40 projector

@Sethnuzum Oh wow, expensive! Been playing with Lazy Lighting on iOS and wanted to move to a desktop more easier interaction with the layout. Wonder if Claude can build something with Three.js...

@chenosaurus Mad Mapper

I set Clawdbot loose on a Mac mini & a Polymarket account with $100. So far, it has lost $15 and spent $20 on Anthropic tokens.

Clawdbot, please send Delta Force and eliminate my enemies.

We learn to speak before we learn to read. Voice is the most natural interface we have. We just raised a $100M to make building voice AI as easy as a web app.

Remembering how my mom got the hot tip about Covid from WeChat before everyone else here knew.

@varoomer @pollenrobotics @huggingface @livekit Haven’t published it yet. Will try to wrap it this weekend!

@chenosaurus @pollenrobotics @huggingface @livekit is there a repo or link?

Gave @pollenrobotics @huggingface Reachy voice, sight, and motion with a @livekit agent. It feels alive. 🤖👻

@johnheywood @ellenjdasilva There was a event last month in NYC with a bunch of local robotics startups. Would be a good list to look at. luma.com/event/evt-0nL8…

@ellenjdasilva @chenosaurus any ideas ?

Are there any amazing robotics companies based in NYC that want to help me throw a sick birthday party for my 7 year old in April?

@Thom_Wolf @pollenrobotics @huggingface @livekit Will do!

@chenosaurus @pollenrobotics @huggingface @livekit You should make a Reachy dashboard app from it and share it on the hub!