Billy Zelsnack 🤖/acc (@zobotics) header
Billy Zelsnack 🤖/acc (@zobotics) profile picture

Billy Zelsnack 🤖/acc

@zobotics

Humanoid labor too cheap to meter.

 Pittsburgh
zobotics.com
Joined April 2022
  • Tweets3K
  • Following2.8K
  • Followers4.8K
510Media
Photo by Billy Zelsnack 🤖/acc (@zobotics)Photo by Billy Zelsnack 🤖/acc (@zobotics)Photo by Billy Zelsnack 🤖/acc (@zobotics)Photo by Billy Zelsnack 🤖/acc (@zobotics)Photo by Billy Zelsnack 🤖/acc (@zobotics)Photo by Billy Zelsnack 🤖/acc (@zobotics)Photo by Billy Zelsnack 🤖/acc (@zobotics)Photo by Billy Zelsnack 🤖/acc (@zobotics)Photo by Billy Zelsnack 🤖/acc (@zobotics)Photo by Billy Zelsnack 🤖/acc (@zobotics)
This is my ANA Avatar XPRIZE robot from just before finals in October 2022. It's mostly 3D printed and cost about $2000. I still don't understand why in 2025 I still regularly come across expensive robots using commercial arms that don't perform as well.
19
22
3
327
49,971
Billy Zelsnack 🤖/acc retweeted
MuJoCo (@mujoco) profile picture
MuJoCo@mujoco3h
MuJoCo 4.0 is out! • Fast or thin objects no longer pass through each other • Soft bodies can now deform permanently and rods can twist • MuJoCo Studio, the successor to simulate, with physically based rendering via Filament Try it live: mujoco.org
22
5
86
4,440
Desktop 3060 12GB is sidelined and now on my laptop 3060 6GB. Buying before was better, but now/BF is prob better than waiting. Not scratch building. 5070Ti is the only reasonable choice? CPU don't care? Only care about non-idiotic ram stick(s) sizing. Thoughts?
1
223
Grokbot WAS so good. Downgraded from Super back to Premium. The cancellation signal to Elon is going to be extreme.
Image from Billy Zelsnack 🤖/acc (@zobotics)
61
19
6
338
16,944
Much easier to see how poor the physical tracking is when I am doing the motions live. I'm going to record some motion data and see how/if it improves. frax retargeter top-left. supermimic trained sim bottom-right. rebocap app right.
4
1
14
1,348
Here's 10000 iterations. Top-left is retargeting. Bottom-left is sim. Right is @rebocap app replaying an animation it comets with. I'll try live trackers tomorrow and record some of my own data to train more with.
Billy Zelsnack 🤖/acc (@zobotics) profile pictureBilly Zelsnack 🤖/acc
@zoboticsNov 5
Trying out the frax retargeter + supermimic tonight. Only at 3500 iterations but it is crazy that I can even think to train this on my 3060 laptop. Excited to see it when I get up tomorrow after it reaches 10000. If it looks good I'll put the trackers on.
1
16
786
Trying out the frax retargeter + supermimic tonight. Only at 3500 iterations but it is crazy that I can even think to train this on my 3060 laptop. Excited to see it when I get up tomorrow after it reaches 10000. If it looks good I'll put the trackers on.
6
4
2
50
3,077
Watching it back I realize that I should have mouse dragged on it to show it was balancing.
1
123
frax teapawt.net/danielpmorton/status/2…
Daniel Morton ✈️ CoRL2026 (@danielpmorton) profile pictureDaniel Morton ✈️ CoRL2026@danielpmortonNov 4
Humanoid retargeting and IK in microseconds [no GPU required] Orders of magnitude speedups, with perfect contact handling: accelerate your robots today at github.com/StanfordASL/frax w/ @drmapavone
2
154
Billy Zelsnack 🤖/acc retweeted
REK (@REK) profile picture
REK
@REKNov 4
We liquidating the humanoid robots we never fought (in stock in SF): 1 Booster K1 2 Unitree R1 basic 3 Unitree R1 EDU 1 AGIbot X2 Ultra 5 Unitree G1-E EDU 2 EngineAI PM01 1 AGIbot A3 2 Unitree H1-E EDU 1 Unitree H2 EDU Match any best price, please DM or respond asap, can ship
14
5
2
70
7,220
AI pause is what is on the table now. We need to move faster. Jan 20, 2028 executive order campaign promise incoming.
Stocko 🦾 (@_Stocko_) profile pictureStocko 🦾
@_Stocko_Nov 4
well, the elections are over and the blue playbook is clear. humanoid robots are off the table. for the next two years it's industrial arms or drones. no humanoids unless it's a sexbot
1
6
759
Saw a El-Sayed interview where he was saying China is not a problem because we pull them along and they can't make progress without us. This insane delusion will be the bad liberals counter-argument to China.
1
113
Using all 15 @rebocap trackers here with Anny MakeHuman. I think that it works quite well. No headset/controllers because their vrmode hijacks SteamVR and I use Virtual Desktop. I have my own vrmode but it's more difficult to show what's going in VR.
2
3
25
2,155
Anny MakeHuman really is not the point, it is just fun to play with the sliders. Lots of next steps. Maybe I'll get the robot retargeting working again.
165
Wait a minute. When did this happen? We really do live in the retro-future now.
Image from Billy Zelsnack 🤖/acc (@zobotics)
1
2
493
I didn't even ask it for an explode view! Flow graph foundation is still not falling apart under its weight. Working on random bits over the entire company tree. Work here is starting with existing motors and then will work back out so that I can plug in custom specifications.
1
3
473
I never edit the nodes or wiring other than manually moving nodes around. Models understand things well enough that I can talk about general changes and never get into the weeds. “let’s make a new component called outrunner motor that we make specializations of our various motors that we can buy”
121
New? Humanoid labor too cheap to meter.
Peter H. Diamandis, MD (@PeterDiamandis) profile picturePeter H. Diamandis, MD
@PeterDiamandisNov 2
The new scarcity is having a purpose and mission big enough to make use of the SI.
1
3
479
A gallon of milk is the SI unit of home humanoid robots.
Jay@Proception (@JayLiStanford) profile pictureJay@Proception
@JayLiStanfordNov 2
"Ain't no way that's a real image." Fair challenge. Here's the raw video: ProHand Gen 2 lifting a full gallon of milk by the handle. One take, 1× speed, teleoperated.
Image from Jay@Proception (@JayLiStanford)
3
18
1,803
Billy Zelsnack 🤖/acc retweeted
Jia-Bin Huang (@jbhuang0604) profile picture
Jia-Bin Huang
@jbhuang0604Oct 30
I'm building a playlist/course for LLM training and inference (WIP). youtube.com/playlist?list=PL… So far, we have common topics like distributed training, fine-tuning, RL, OPD, serving, spec. decoding, and basic optimizers. What else should I include to make this more complete? 🤔
AI Training and Inference

AI Training and Inference

This course covers fundamental training and inference techniques for Foundation Models (e.g., LLMs, VLMs).

youtube.com
34
96
1
1,083
53,615
Waffle/Waffle Maker couples costume turned out pretty well.
Image from Billy Zelsnack 🤖/acc (@zobotics)
1
3
383
Grok helped come up with the idea. grok.com/imagine/post/969e3e…
🤖 Made with AI
203
Load more