Eureka! NVIDIA Research Breakthrough Puts New Spin on Robot Learning

[bnew]

Spoiler: thread continued

Spoiler: full text & large images

1/12
@RemiCadene
Meet the game-changer: SO-100

Crafted by @therobotstudio and @huggingface

At 1/3 the cost and 2x the capabilities of our previous arms, it's the most accessible, high-performance robotic arm for $115. Easiest DIY at home!

1/ Link and details in thread

2/12
@RemiCadene
2/ Fully integrated with @LeRobotHF, the go-to for tutorials, datasets, and pretrained models — all in python

Now is the best time to get into AI for robotics!

Github: GitHub - huggingface/lerobot: LeRobot: Making AI for Robotics more accessible with end-to-end learning
Discord: Join the LeRobot Discord Server!

3/12
@RemiCadene
3/ Both arms are designed to fit perfectly on a standard 3D printer bed.

To ease assembling, we use the same motors

4/12
@RemiCadene
4/ Mark your calendars for October 26th-27th and join us online to build them together! (Details coming soon)

Order your parts now! GitHub - TheRobotStudio/SO-ARM100: Standard Open Arm 100

5/12
@ramkumarkoppu
any plans to sell assembled arms to use out of the box?



6/12
@RemiCadene
Possibly the off the shelf version. We are still working on it.



7/12
@thejamesvance
Looks awesome. When will capability specs be shared?



8/12
@RemiCadene
what specs do you want to know?



9/12
@marten_wikman
Is it possible to use the waveshare ST3215 servos to build this?



10/12
@RemiCadene
I dont know, but it looks like repackaged feetech motors :O



11/12
@sog_on_bird_app
I just printed the Koch arm aaaaaa



12/12
@RemiCadene
haha don't worry the koch arm is really good as well ! but we felt a more affordable arm was possible




To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

 

[bnew]

 

[bnew]

Spoiler: thread continued

Spoiler: full text & large images

1/11
@TheHumanoidHub
I had the privilege of touring Figure's Sunnyvale office yesterday, though calling it just an office would be a massive understatement. There's a whole lot more happening under that roof than you'd expect.

The first thing that struck me was how densely packed the space is. Around 140 employees work in an open, collaborative area with minimal partitions. It’s clear why they’re moving to a much larger facility to accommodate their growing team and scale up production. The new facility is going to be eight times larger.

Brett gave me a full tour, and the energy in the place was palpable. In the test area, I saw a mockup station that simulates a real BMW workstation where an F.02 robot was practicing a dexterous pick-and-place task, while another one, behind curtains, was using OpenAI’s model to handle everyday objects.

The production area was equally impressive, with four assembly stations where robots were in different stages of assembly. Brett mentioned they’re currently building around 2-4 robots per week.

What’s really remarkable is how many components are made and assembled on-site, from outer shells and wiring harnesses to control systems, actuator parts, fingers, and even the packs that house batteries and the computer. Dyno-testing, calibration, and system-integration testing all happen under one roof.

This level of integration requires a full team effort. Designers, software/AI engineers, mechanical and electrical engineers, and robot trainers all work together in harmony. As someone who comes from a software-only domain, I left in awe of the operational complexity required to build a working robot.

On the financial front, Brett confidently shared that they’ve got the cash burn under control. At their current rate, they could theoretically operate for 14 years on existing funding, though of course, that will change as manufacturing and operations scale.

I also got a sneak peek at their next-gen robot, F.03's design. It’s incredible to see how far the company has come in just 2.5 years, going from nothing to Figure 01, to F.02, and now developing F.03. If they continue at this pace, the next few years are going to be incredibly exciting.

Watching a humanoid robot walk around and perform tasks was nothing short of mind-blowing. The future felt a lot closer and more tangible from inside their walls. Huge thanks to @adcock_brett for the tour.



https://video.twimg.com/ext_tw_video/1841538240165380096/pu/vid/avc1/720x720/YO1D1j3yQu_7CECN.mp4

2/11
@jmfinvestments
What are your thoughts on their ability to scale?



3/11
@TheHumanoidHub
I'm no expert in manufacturing, but I imagine they’ve gained some valuable lessons from small-scale production



4/11
@_FreddyHS
Why is the walk so slow? Is the Optimus robot just on another level in terms of walking speed?



5/11
@TheHumanoidHub
Brett said that completing long-horizon tasks reliably is a higher engineering priority right now. They'll get close to human walking speed by the end of this year.



6/11
@FabienMusty
Great! I suppose that sequence is fully autonomous? Did they elaborate about the software? Any default OS behavior cooking to blend the trained sequences seamlessly and pick and choose elements to obey demands?



7/11
@TheHumanoidHub
Yes it is autonomous. I don't have a good answer for the latter, I'm guessing the whole sequence is trained as a single task end-to-end.



8/11
@doctorBobG
Which version is shown in the video?



9/11
@TheHumanoidHub
That's F.02.

F.03 is under development.



10/11
@CJDGiesen
"On the financial front, Brett confidently shared that they’ve got the cash burn under control. At their current rate, they could theoretically operate for 14 years on existing funding, though of course, that will change as manufacturing and operations scale."
NO. That is only true if they can use Compute resources for free. If they need to build out a Tesla Cortex-like Supercompute Cluster, they are broke in 2 weeks. Don't be fooled. This is a Hyperscalers game.



11/11
@TheHumanoidHub
Good point.
They may not have to do all hyperscaling in-house. They're partnering with OpenAI on the foundation model of general-purpose robotics. The more general the foundation model is, the less data-intensive the policy network would be.




To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

 

[bnew]

Spoiler: thread continued

Spoiler: full text & large images

1/11
@TheHumanoidHub
LIMX Dynamics is now taking pre-orders for TRON 1 [price $15k]
⦿ Easy transition between bipedal and wheeled
⦿ Remote-controlled walking out-of-the-box
⦿ Open SDK & interface support for Sim2Real deployment
⦿ Peripheral compatibility with LiDAR, cameras, and robotic arms



2/11
@Bacheeta24
What will be the use case for it ?



3/11
@TheHumanoidHub
Same as a robot dog - and the skills and behaviors easily transfers to a full humanoid.



4/11
@Brenten55
That’s cool . It might be the answer - wheels on/wheels off. I’m excited with these AI companies working on Robots!



5/11
@GlueNet
Modular design?

Be still my heart, these robots are advancing at DO NOT KICK FRIEND ROBOT!



6/11
@kiranadimatyam
Is it for industrial use? I do not see the reason to use it at home.



7/11
@AI_TechnoKing
This little guy should be in the robot Olympics!



8/11
@BenPielstick
It has arms?



9/11
@JohnFSDBull
Stick a seat on that and I'll buy it



10/11
@Jesusnumber_2
Is that fukking claptrap



11/11
@WealthArchives
@ServeRobotics y'all see this shyt?




To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

 

[bnew]

Spoiler: thread continued

Spoiler: full text & large images

1/11
@GoingBallistic5
Gen 1/2 Wrist vs. Gen 3 Wrist

At We, Robot we got a glimpse of the new 23 DoF (yes 23, not 22) hand.

In this post I will only address the change in actuation strategy for the wrist. Later posts will address the tendons.

The Gen 3 wrist is an improvement over the Gen 1/2 design. It is still driven by linear actuators. However, they solved an issue with the original design.

In the Gen 1/2 wrist (left), the black actuator needs to rotate to prevent the piston from binding during travel. It is a three-bar mechanism.

In the new wrist (right) an additional linkage is added to allow the actuator to remain stationary. Technically this is a four-bar mechanism.

The trade-off is an additional part is needed to keep the actuator fixed. This is needed to make the servo stack of the forearm compact, so the wrist actuators can be easily integrated into the design.

In the Gen 3 design, the new linkage is not straight as in the animation, but crooked to allow for better integration.

This new mechanism can be seen in video clips like this taken at We, Robot .



https://video.twimg.com/ext_tw_video/1846399022451314688/pu/vid/avc1/974x720/sluEaKPeZ5b3vOjO.mp4

https://video-ft.twimg.com/ext_tw_video/1846399085567266816/pu/vid/avc1/512x570/wNEaoTl2c3CX9pqg.mp4

2/11
@a_meta4
Would love to chat with you in a space if you have time!



3/11
@GoingBallistic5
Absolutely



4/11
@VacekvVita
Thanks for your explanation!



5/11
@seti_park
Could this technology be relevant?

[Quoted tweet]
Advanced IK Features and Target Handling

Disney's IK solver incorporates sophisticated features for handling complex scenarios and maintaining robot balance.

IK Target Handling (Figure 6A, 6B)
Safeguards against unreachable targets
Finds closest feasible solution for out-of-reach targets
Incorporates frame objectives for position and orientation control

Center of Mass (CoM) Control (Figure 6C)
Implements CoM objectives for balance maintenance
Crucial for stable bipedal walking and complex movements
Allows for quasi-static approach in slow walking scenarios

Key Advantages
Enables precise end-effector control while maintaining overall stability
Facilitates creation of more natural and expressive robotic movements
Applicable to a wide range of robotic structures, including those with kinematic loops

This advanced approach ensures that the robot can perform complex movements while staying within its physical limitations and maintaining balance.

6/11
@Able2Xplore
why limit to human capability? i imagined they would have all range and why not?



7/11
@TheeMsMolly
Scott I need a very particular set of skills



8/11
@M_IsForMachine
I remember the wide assortment of four-bar linkages on display in the engineering buildings at Cornell.

They seemed pretty old, say from the 1930s, but perhaps still too new for you to have seen them there.



9/11
@stevenuecke
Interesting! Looks like some off axis forces were added as part of this change, which may have required some additional modifications to the actuator.



10/11
@long_elon
Did you get confirmation from someone at Tesla of it being 23 DoF or is that number from you counting the DoF yourself?



11/11
@DeCochesei
?




To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

 

[bnew]

Spoiler: thread continued

Spoiler: full text & large images

1/8
@agilityrobotics
We recently had @CNBC in the house for a tour of RoboFab, the world's first humanoid robot factory.

There were already more Digit's in the world than any other humanoid robot, and that number is getting much larger.



https://video.twimg.com/ext_tw_video/1844484208619814913/pu/vid/avc1/1280x720/8Lipm7ZeYK8_P0xz.mp4

2/8
@OrniasDMF
Tbh when you skip things like hands and rely on QR codes, it's easy to enter the market first.
The companies that are aiming to perfect every element may enter later but anything less will be unable to compete.
Sanctuary is the same with their lack of legs.
Tesla will overtake.



3/8
@agilityrobotics
At a certain point, you can't continue to chase perfection in a lab. You have to get into commercial environments and deliver a product.



4/8
@WiBotic
We love Digit! Can't wait to see the compatibility with human workers coming next.



5/8
@chrismatthieu
I Digit



6/8
@AlgarnNFT
I find the intersection of AI and robotics fascinating. What's the potential impact of mass-producing humanoid robots like Digit on industries and society?



7/8
@i_bioloid
Very cool is this the full video?



8/8
@hackertwinz
So cool! Can't wait to build a Tripedal Robot factory one day

[Quoted tweet]
Here is a sneak peak of our next generation walking robot, Tripedal! It has so many improvements, when its walking we'll publish a deep dive video. Its fully 3D printed and uses 10 Servo Motors


https://video.twimg.com/ext_tw_video/1843008963052232704/pu/vid/avc1/720x1280/nvwyd0U9W4W6z7q8.mp4


To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

 

[bnew]

Spoiler: thread continued

Spoiler: full text & large images

1/7
@TheHumanoidHub
New video by RobotEra.

XBot-L becomes the first humanoid robot to walk on the Great Wall of China.

[Quoted tweet]
New humanoid robot startups are popping up like popcorn.

A Chinese startup, RobotEra, just released a new demo showcasing new hands designed for their robot XBot (XiaoXing). The demo makes no claim of the robot being autonomous; it's probably teleoperated.


2/7
@MrDanielBar
What would it do if it saw a Mongol army riding towards it?



3/7
@TheHumanoidHub
I'll have my Chu Ko Nu's and Skirmishers ready.

4/7
@RottenHeadRed
Meh



5/7
@ByeonChansoo
walks more human than @JoeBiden



6/7
@XTeamPal




7/7
@Renata_Stochmal
The robot should have better shoes to walk on this kind of surface! I found these for Optimus, they fit this robot too …and if you could equip them with FSD, wow! The robot would be happy!

[Quoted tweet]
Walking is fine, but eventually he'll have to do some work too



To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

 

[bnew]

Spoiler: full text & large images

1/7
@TheHumanoidHub
New demo of Chinese company UBTECH shows Walker S1 carrying a 36 lb (16.3kg) tote while walking on a treadmill.

"The integrated perception & control, along with the end-to-end learning-based whole-body control, enable the bot to achieve dexterous manipulation & stable walking."



https://video.twimg.com/ext_tw_video/1846220713575043073/pu/vid/avc1/720x720/-hELlhGsywTES3mJ.mp4

2/7
@paulctan
Control system/neural net needs work. Too much oscillation overshoots going on there.



3/7
@Prashant_1722
Robo era is dawning upon us



4/7
@JohnFSDBull
I'm pretty sure people can do that and more - literally everything else. I'm skeptical when these things will actually be more useful in most situations



5/7
@dugwool
We have yet to see humanoid bots in a real working environment from anybody. Leads me to think that there is a way to go yet. Similarly LLM's. Same problem - how to internalize a world. Without consciousness.



6/7
@ChongZitaZhang
This is clearly model based planning rather than end to end learning based control



7/7
@TechRenamed
Ok but does it talk




To post tweets in this format, more info here: https://www.thecoli.com/threads/tips-and-tricks-for-posting-the-coli-megathread.984734/post-52211196

 

[bnew]

 

[bnew]

 

[bnew]

 

[bnew]

 

[bnew]

 

[Micky Mikey]

where can i get one of these?

 

[bnew]