Bernt Bornich
@BerntBornich
Followers
2,589
Following
46
Media
12
Statuses
75
CEO and founder of @1x_tech (Androids and Embodied AI)
San Fransisco
Joined March 2022
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
LOVTITUDE FANMEETING
• 644626 Tweets
namjoon
• 613710 Tweets
FB MADE YOU BLUSH
• 497305 Tweets
Doge
• 302412 Tweets
Xavi
• 277749 Tweets
Kabosu
• 198965 Tweets
Flick
• 147115 Tweets
#한성수_방시혁이랑_손잡고_한강가
• 145662 Tweets
The ICJ
• 132717 Tweets
クロリンデ
• 99193 Tweets
LEAVE SEVENTEEN ALONE
• 95769 Tweets
シグウィン
• 92661 Tweets
Laporta
• 86132 Tweets
Ten Hag
• 85848 Tweets
ノーヒットノーラン
• 72290 Tweets
Memorial Day
• 54242 Tweets
#ساعه_استجابه
• 45465 Tweets
アルハイゼン
• 43830 Tweets
Leclerc
• 41907 Tweets
Congratulations LISA
• 41718 Tweets
McKenna
• 41335 Tweets
#ガルクラ
• 36322 Tweets
العدل الدوليه
• 35359 Tweets
#SRHvsRR
• 31042 Tweets
تشافي
• 28123 Tweets
Coutinho
• 26949 Tweets
INEOS
• 21918 Tweets
Super Size Me
• 21210 Tweets
Morgan Spurlock
• 20680 Tweets
Knives Out
• 16481 Tweets
Wake Up Dead Man
• 13890 Tweets
Benoit Blanc
• 12270 Tweets
When your colleague asks to bring an old robot home for the weekend and then sends you this
8
74
539
First batch of droids at the new studio🦾 Major update on scaling our end2end model going live in 3 weeks
42
67
357
New progress update on the droids dropping in 4 weeks, looks like Moravec's paradox might be false and we just didn't have the data
11
31
257
Lot’s of great hype about the large multimodal models right now. But how do we get to trillions of tokens for embodied actions? My bet is VR-Teleoperation and shared autonomy.
13
26
158
NEO just picked its first cup, excited to finally share some hands-on details🧵
6
19
129
20 years ago this year ASIMO 2004 launched and is still seemingly state of the art. Why did it fail?
🧵The Bandwidth Dilemma: Understanding Human Biomechanics to Build Better Robots
2
18
110
Up next is training our droids to tidy up the office and handle logistics and material in our production line for EVE and NEO. All with exactly the same data collection + training pipeline we ship and use in this video. General-purpose autonomy affords vertical integration
5
11
75
Latest
@1x_tech
software update with contact-rich manipulation skills and general picking
With all capabilities merged into a single set of goal-conditioned network weights, we see first signs of zero-shot behavior for manipulation tasks
1X’s mission is to create an abundant supply of physical labor through androids that work alongside humans. We're excited to share our latest progress on teaching EVEs general-purpose skills. The following is all autonomous, all 1X speed, all controlled with a single set of
60
159
1K
8
7
74
I've been asked about my 2024 predictions in AI/Robotics
1) Humanoid companies bifurcate into industrial/structured and unstructured/human-centric based on hardware directions
2) We finally see generalization in RoboticsAI
3) Signs of embodied data significantly improving LLMs
🧵
3
11
68
I'm pleased to announce the closing of 1X Series A2 led by OpenAI.
We announce today the successful close of our Series A2 funding round.
We intend to use the funds to increase the efforts of building our upcoming bipedal android model NEO.
12
40
131
4
6
49
Proud to share that
@1x_tech
has closed our Series B.
We’re taking the next step toward our vision: Androids in every home, enhancing daily life. The team at 1X makes these milestones possible, and we can’t wait for you to bring an NEO home.
3
4
49
20 degrees of freedom
Grip strength enough for NEO to deadlift 70kg
Tendon-driven with force-control
Naturally compliant and safe
Designed for 1.000 falls between failure
True skill transfer from human experts requires matching dynamics & kinematics between droids and operators
0
2
35
Most comprehensive overview of the field I've seen so far! Really heating up🔥
1
7
35
Bonus:
How fast are we humans really?
-Visual Reflexes: 150-250 ms (5hz)
-Auditory Reflexes: 140-160 ms (6.7hz)
-Touch Reflexes: 100-150 ms (10hz)
-Local reflex (muscle bandwidth): 25 ms (40hz)
Effective bandwidth for controlling touch/collisions:
100-250 ms (4-10hz)
0
1
30
Bonus: 2024 will be the year most of the tech world realizes how humanoids will completely change the economy; we will all disagree on timelines but agree it will happen. This will result in a similar craze as LLMs in 2023.
1
1
24
As AI’s commoditization of software accelerates, the real winners will be companies that combine hardware IP with unique data moats. The value is shifting from software to hardware and data.
0
1
20
Halodi is now 1X! After 7 years of vertical integration in motor designs, actuators, sensors, and AI, our android Eve is rolling off the line. This month, as part of building the brand, we will start revealing what's next and what's been brewing in the lab for the last two years.
0
2
19
Honored to present to such a knowledgable crowd at Mars 2023
0
2
18
The funds will be used to increase our manufacturing in Norway and Dallas
2
4
14
1X designs our androids with tendons, loosly inspired by human muscles, based on our actuators that are now 80% the density of muscle. Intelligence lies as much in the dynamics of our bodies as in our brain.
1
0
15
And, of course, enhance the next iteration of our bipedal android NEO.
1
2
11
Kinetic Energy is 1/2*(m*v^2+I*w^2). Notice the velocities are squared. That's why, in nature, nothing ever moves faster than it has to. ASIMO has motors with 100:1 reducers, rotating 100x the joint speeds, that's 10.000 times the energy. Think of a big flywheel spinning up
1
0
12
Animals are masters of adaptability and compliance. We are purpose-built to move fast with minimum energy so we can collide safely. All we do is collisions, picking up a cup, taking a step, grabbing that door handle, and we are not fast enough to react to sense of touch and stop
1
0
9
In the words of our master Bruce Lee, 'Be as water, my friend.' Water takes the shape of anything it enters. - Whatever situation life puts you in, embrace it and adapt - Don't remain stiff.
NEOs feet have "kettlebell inertias of 0.2kg, 100x less, but Bruce still wins with 0!
1
0
8
Signs that LLMs greatly improve with the same embodied data (how actions changes the state of the world, modalities like egocentric touch/audio/vision). But no one is yet training foundation models on this combined data due to the large players staying clear of hardware
2
0
8
We started at the wrong end when autonomous cars became the bridge from digital to physical. I’m stoked to see how the coming wave of humanoids will leverage not being a 2.5tonn death machine
0
0
8
ASIMO can do tiny jumps on 1 foot and have a small flight phase when running, an immense feat of engineering given the amount of energy that needs to be disipated at every hop, the "flywheel needs to instantly stop" when the foot hits.
1
1
8
@Joemf1997
Our EVE android in this video is already being produced and deployed with some large enterprises. You might randomly stumble upon one soon in the wild
0
1
7
💯
Robotics is a field with lots of historical burdens. Too many poorly engineered software, algorithmic patches, and outdated mindsets.
Foundation models will happen slowly, then all at once before you know it, burying these legacy stuff once and for all. Just like how ChatGPT
41
81
576
0
0
7
Compliant/soft systems will leverage the advantage of safe manipulation and efficient operation in unstructured environments to gather more diverse data (consumer and services in addition to the industrial). The jury is still out on if this is required, but this is our bet at 1X
1
0
7
Stiff humanoids (most of them) will encounter problems deploying safe manipulation and operating in the messy uncertainty of the world but find applications in structured environments where calibration and adaptation of the environment are feasible (factories/logistics)
2
0
7
Hats of to Honda who perfected this approach to robotics way ahead of its time. Sadly Honda discontinued ASIMO in 2018, after producing around 100 units, it never managed to leave the lab.
2
1
8
Playing with human demonstrations in VR, hopefully soon easier to teach the robot to tidy than my kids 😁
0
2
7
You need a body that has human-like dynamics to be able to approach AGI. Reality is to complex to be fully represented in a book
0
1
6
Narrow wedge approaches to LLMs never worked and neither will they for humanoids, that's why safety and cost is king. Maximize the width of your data distribution and train on your test set when you can.
0
0
7
Manipulation has not yet seen generalization like Pix2Pix/LLM (RT-X while impressive was a negative result on increasing dexterity from non-robot data). Deployments at scale will enable generalization through large-scale immitation learning (For 1X, NEOs expansion into consumer)
1
1
6
So when ASIMO is running, we can compute the rough equivalent inertia of the spinning flywheel to its foot. It works out to about 20kg. So it's running around with a 20kg kettlebell strapped to each foot (why robots walk weird and are unsafe) and roughly 10kg to each hand.
2
0
5
@GoingBallistic5
Do androids dream of a secret place where they spill their coffee, studying fluid mechanics and answering the great question, why do humans choose to spill their morning coffee?
0
0
5
Cruising through LA with a VR headset, hunting for that high-speed connection to link up with my droid in Europe for the gig — this is peak cyberpunk surrealism!
0
0
4
@Joe__Black__
@ericjang11
@RoboDepot
@elonmusk
@aakamisaka
@1x_tech
@adcock_brett
@DrJimFan
@_milankovac_
@julianibarz
@realgeordierose
Months, not years, not weeks, it's worth the wait
1
1
5
Keep hearing robotics hardware is already decades ahead of the AI to use it. I think our (Halodi's) most significant advantage is realizing how wrong this is.
1
0
5
The potential with this partnership is what makes me most excited. Exploring how AI can be embodied in our androids
1
0
4
What if LLM's and LDM's main contribution becomes teaching the general public source criticism... It could be one of humanity's most tremendous leaps forward.
0
0
3
@JoannotFovea
You won't need to wait for long, just got a EVE in my house for Christmas, gonna run a kind a daily video blog here when we get started (it's our industrial version, not the new NEO thats meant for the home, but will be great data and learnings
3
0
4
As an 11-year-old I decided I would go to Japan and work with humanoid robots when I grew up. Halodi in Norway, Tesla, Agility, Figure, ... all in the US, what happened?
1
0
4
Not a problem right? We can just be fast enough to stop when touching something? Well, if we move slow or know where everything is we kinda can. This works in the lab, but the real world is hard and unpredictable.
1
0
4
Embrace the grind, just make sure there's purpose at the end, we're not drones
0
1
3
@atroyn
@HalodiRobotics
Eve has 3 unique motors. Our actuators are more part of the structure as we use cable drives, so they differ a bit from limb to limb, the same as other mechanical parts in the robot. For our next-gen biped, we have increased to 5, including hands (which would put Tesla at 7)
0
0
2
@Joe__Black__
@ericjang11
@RoboDepot
@elonmusk
@aakamisaka
@1x_tech
@adcock_brett
@DrJimFan
@_milankovac_
@julianibarz
@realgeordierose
More info here we're all in on full general purpose with compliance and safety
NEO just picked its first cup, excited to finally share some hands-on details🧵
6
19
129
1
0
2
Great summary of the week
0
0
2
@loybeek
@EngNadeau
@soumithchintala
People vastly underestimate the value of knowledge transfer from humans into our robotics systems. This mandates designing for the same passive dynamics and kinematics as humans. Ask GTP-3 how to unscrew a jar of jam.
0
0
2
@tom_doerr
Contact dynamics and preferred impedance etc is poorly represented in simulation and only indirectly in video. Physical interactions hide alot more complexity than what is obvious on the surface.
1
0
2
@andrewkirima
@EngNadeau
@HalodiRobotics
Really enjoyed our conversation
@andrewkirima
, not often do I meet someone truly working on the zero to one long-bets that nudge the world forwards. PS! it's never a bad time to talk about motors ;)
0
0
2
@Joemf1997
Droids are all awake and working hard on the crunch to make a splash. Just a few days left
0
0
1
@jamesgiammona
Not sure I have any answer. My guess is there is more out there than we know. Alternatively something other than just funding is hindering commercialization of their research.
0
0
1
@levitimg
Footprint while working with large forces in a big workspace for one, and ofcourse easier to teach through imitation learning. That goes for all humanoids
0
0
1
@andrewkirima
Peter Barret in Playground Global that triggered this rant. But even Elon/Tesla claims this is mostly an AI problem: "just need a robot that can kinematically fit in our space"... Truth is complexity of the problem is inverse to how natural the dynamics of the machine is.
1
0
1
@chris_j_paxton
Not in anyway straightforward. Huge challenges to overcome with respect to privacy and legal. Not to mention safety. But as we are increasingly able to deploy droids with base autonomy+teleop I think it has a path to collecting the diversity of data needed.
0
0
1
@roboticsPundit
I'm talking about the passive dynamics, we can mask parts of the inertia if we have fast enough reaction time (bandwidth) and enough motor torque. But in reality this breaks down pretty fast and you see huge impacts or slow movements
1
0
0