echo

171 posts

echo banner
echo

echo

@EchoRobots

trying to build a new representation of a 3d world. sharing the journey

Katılım Mart 2026
228 Takip Edilen46 Takipçiler
Sabitlenmiş Tweet
echo
echo@EchoRobots·
we keep treating robotics as a smarter-policy problem. but watching what actually fails, it's almost never policy. it's description - the robot's understanding of what's in front of it. shape, identity, affordance, intent. we're building better hands while drawing the world in stick figures. control is mostly solved. description is the open problem nobody is treating as one.
English
1
2
9
709
echo
echo@EchoRobots·
@urieli17 הגיע הזמן להכניס את הקהילה לרובטיקה
עברית
0
0
0
23
Uri Eliabayev
Uri Eliabayev@urieli17·
יש עכשיו דיבור על מבחן ביצועים חדש למודלי AI שסוף סוף מאתגר אותם. כולי מתלהב ורוצה לחקור אותו לעומק ומחפש איך ללמוד עליו עוד ואז נכנס לקהילה שלי ורואה שאחד המחברים שלו הוא ישראלי והוא העלה על זה פוסט. כל פעם מדהים אותי מחדש אילו אנשים מדהימים נמצאים אצלי בקהילה ולא ידעתי על זה
Kilian Lieret@KLieret

The first ProgramBench task was just solved by GPT 5.5 high/xhigh. Interestingly, high/xhigh picked two different languages for the task (C vs Python). GPT 5.5 xhigh was significantly better than Opus 4.7 xhigh in all metrics. 🧵

עברית
3
1
107
11.7K
echo
echo@EchoRobots·
@crypt0lake not if you tell him to do a good job in the end
English
0
0
1
40
cryptolake
cryptolake@crypt0lake·
multiple agents being the same model but different prompts is overly useless btw
English
6
0
19
1K
echo
echo@EchoRobots·
trying to solve it with a new file format. no urdf, no mjcf, no yaml. no more-0.683 , just intent and python. still early.
English
0
0
1
18
echo
echo@EchoRobots·
robotics still writes in assembly. CAD → URDF → MJCF → endless YAML, magic numbers like -0.683 derived from nowhere, all to put one apple on one table. ML had its python moment twenty years ago. robotics still types coordinates by hand. this is just software, before AI enters the picture and add its magic. someone needs to build the python for this layer.
English
1
1
4
3K
echo
echo@EchoRobots·
our try at the utah teapot, remade in echo. came out closer to aladdin lamp than teapot. taking it.
echo tweet media
English
0
0
0
16
echo
echo@EchoRobots·
This is the most famous 3D object ever made. You've seen it a thousand times. Nobody told you its story. 1975. Martin Newell, PhD student at Utah. He needs a curved object to test his renderer. His wife points at the kitchen counter. Their Melitta teapot. He sketches it on graph paper. 28 Bezier patches. Smooth surfaces, each one a single equation. Pure math. 50 years later, every GPU on earth still has to chop those patches into thousands of flat triangles before it can draw them. Triangles are the only shape modern hardware knows. The pioneers had answers we abandoned for cheaper hardware in 1996, and never went back to check. Maybe the future is hiding in the past we forgot to read.
echo tweet media
English
1
0
1
35
Stocko 🦾
Stocko 🦾@_Stocko_·
@EchoRobots i guess another way to phrase it is poor societies can't afford to build hardware
English
1
0
1
17
Stocko 🦾
Stocko 🦾@_Stocko_·
poor societies write software rich societies build hardware
English
3
0
11
429
echo
echo@EchoRobots·
@_Stocko_ the deeper i go the more these axes blur. each carries the other's answers.
English
1
0
1
18
echo
echo@EchoRobots·
@chris_j_paxton love your blog already read this on. where does the non-predictive kind fit , typed graph + physics, no rollout. world model in your view ?
English
0
0
1
43
Chris Paxton
Chris Paxton@chris_j_paxton·
I'm not sure id consider any "robot foundation model" company a world models co, unless its also training world models... The way people use terms changes and thats fine. Silly to be upset about eg world labs being included because they dont learn a conditional dynamics model. But some of these just arent world models at all It is however the hot thing right now and the post is mostly a good overview imo
Deedy@deedydas

This is the single best read on World Models and one of the most important reads in AI. $10B has flowed into "world models" in the last 18mos, from Yann LeCun to FeiFei Li. The promise is, like LLMs, world models will provide the data it takes to scale robotics foundation models, and solve robotics. ..but the word has been abused to mean one of many things. This post unpacks: – What 5 traits makes a world model? – How do the different approaches stack up? – What is it used for within and beyond robotics? – Where is the opportunity? – Citations to research, news and blog posts Companies / products in the space include: – BigCo products: Google Genie, Tesla Optimus, Nvidia DreamDojo, DreamZero, Microsoft Muse – Pure world model: AMI Labs, World Labs, Runway, Rhoda, Decart, Spaitial, Odyssey, Embo, Dream Labs, OneWorld – Robot foundation model cos: Skild, Physical Intelligence, Figure, Mind Very likely one of the seminal technologies of the next decade.

English
2
4
36
4.8K
echo
echo@EchoRobots·
world models are CAD systems in disguise.
English
0
0
2
42
echo
echo@EchoRobots·
The difference between a robotics researcher and a game developer gets smaller every year.
English
0
0
3
46
echo
echo@EchoRobots·
@_Stocko_ Agree! Someone will do it in the new future I guess
English
1
0
1
21
echo
echo@EchoRobots·
in software, you write code in one language and it runs. in robotics, you define geometry in CAD, retarget to URDF, simulate in MJCF, hand-type coordinates in YAML, and pray the trained policy survives the gap. five formats. one simple secne. the bottleneck isn't AI. it's translation.
English
2
0
6
298
echo
echo@EchoRobots·
@Robosynx_AI You got me there on the physics! But in this lab, it's the opposite,the robot has 'stage focus.' It only chooses to exist in a working state when it knows it's being recorded. 😅
English
1
0
0
62
rpx
rpx@Robosynx_AI·
@EchoRobots Quantum particles: fail the instant you record them Robotics demos: only work if you record them
English
1
0
1
38
echo
echo@EchoRobots·
Robotics demos are like quantum particles. They only work when observed by a camera.
English
1
1
3
105
echo
echo@EchoRobots·
@LightwheelAI deformable manipulation is where the description bottleneck really bites, you can't hand-write a URDF for a cloth that's mid-fold. curious what the eval metric looks like for phase 1: task completion, trajectory smoothness, or something else?
English
0
0
0
21
Lightwheel
Lightwheel@LightwheelAI·
60+ teams have submitted to the LeHome Challenge 2026. We are now in the evaluation stage. Folding clothes is easy for humans. Still brutally hard for robots. That is why LeHome focuses on deformable object manipulation, one of the hardest open problems in robotics. Phase 1 wrapped with teams submitting policies across four folding tasks: long sleeves, short sleeves, pants, and shorts. 16 winners will have Phase 2 awards. The top 8 teams will compete live at ICRA 2026 on June 4. #Robotics #PhysicalAI #EmbodiedAI
English
2
0
4
1.5K
echo
echo@EchoRobots·
In classical ML/DL/AI, you must show generalization so your work will be accepted. For some reason, in robotics you can validate yourself on the train and be rewarded, this must change if we want to get to real value.
Jitendra MALIK@JitendraMalikCV

We have seen some impressive robot manipulation policies recently. This is great, but for these results to be convincing (and practical) we should insist on generalization across 1. Object location (within some range) 2. Different instances of an object category 3. Background clutter. Authors should present experiments which demonstrate the range of variation which can be handled. Far too often the policy doesn't even generalize across all instances of an object category! Legged locomotion policies were convincing only when they worked across different terrain as in ashish-kmr.github.io/rma-legged-rob… We need to do the same for manipulanda.

English
0
0
0
155
echo
echo@EchoRobots·
@0fir0z I'm looking for ways to do better robotics, the real one, not physical ai.
English
1
0
1
18
Ofir Ozeri
Ofir Ozeri@0fir0z·
“Physical AI” is what people that don’t know robotics say when they want the robotics trend to fit neatly into the AI hype cycle. But it’s not only that robotics isn’t a branch of AI. The robotics revolution that’s coming will end up dwarfing every vector of the AI wave.
English
2
0
5
305
echo
echo@EchoRobots·
@0fir0z I am actively exploring it ☺️
English
1
0
1
23
Ofir Ozeri
Ofir Ozeri@0fir0z·
@EchoRobots Yeah, it’s the whole vibe-approach that makes people think that just spray some AI powder on a problem and see it solved. Robotics has had enough of a bad rep for its legacy demo culture, we don’t need this shit
English
1
0
1
55

Keşfet

@urieli17 @crypt0lake @_Stocko_ @chris_j_paxton @elonmusk @BarackObama @taylorswift13 @cristiano