Exploring the Frontier of Physical AI. All opinions are my own.
- Tweets 2,309
- Following 1,911
- Followers 10,895
- Likes 2,517
Pinned Tweet
Our first podcast episode is finally live!
I personally loved the casual chat after the presentation (From 46:05 onwards).
Sorry about the audio—my mic didn't record, so I had to re-record my lines as best as I could remember them.
And huge thanks to @LujieYang0 and @chris_j_paxton.
Looking forward to seeing more research published on our Booster robots 😎
How did the robots at the Spring Festival Gala pull off parkour and backflips?
Check out our Teaser Part 1.
Featuring Lujie Yang @LujieYang0 with her presentation: From Math to Motion: Dynamics, AI and Robotics.
Also huge thanks to @chris_j_paxton for collabing as well. Full episode coming soon.
19
15
129
71,899
31M parameters, 7ms latency, fully on-device.
Good enough to run speech synthesis locally on a robot without sending every response to the cloud.
A new frontier for on-device speech.
We're open-sourcing Phonon, a suite of TTS models under 50m parameters, setting a new standard for naturalness and accuracy across languages, with the broadest inference support to date.
Phonon-31m is available under Apache 2.0.
gradium.ai/phonon🧵
15
911
Motion2MixamoRig turns a single-person video into reusable 3D motion for Mixamo-rigged characters, with Blender and Unity-ready output.
This offers a lower-cost path from human video to motion data for simulation and humanoid training.
github.com/MaxLiu22/Motion2M…
3
2
21
1,957
Back in April, I posted about Nature Will @naturewill when they first caught my attention. Now we finally get to see what they’ve been building.
Their first model Natura-1 is trained on nearly 1M hours of egocentric human data, combining dexterous control, zero-shot instruction following, and long-horizon execution.
If this scales, it points to something much more important than another manipulation demo: learning physical intelligence from human experience, then generalizing zero-shot to new tasks and conditions on real robot hands.
Definitely one of the robotics releases I’ve been waiting for!
Introducing Natura-1, our first general foundation model for the physical world.
It follows natural-language instructions zero-shot and turns them into action with dexterous hands.
Natura-1 is a 50-billion-parameter mixture-of-experts (MoE) model pretrained on one million hours of egocentric human experience.
On top of this foundation, we conduct efficient post-training on a limited amount of robot data to adapt pretrained world knowledge to dexterous manipulation, yielding a model that generalizes to a wide range of unseen scenarios.
Natura-1 follows complex natural-language instructions zero-shot, reasoning through spatial constraints, multi-object relationships, and negative conditions.
It can also autonomously carry out long and continuous workflows to perform long-horizon executions.
Natura-1 provides early evidence that generalization—and even emergent behavior—can coexist with the complexity of dexterous hands.
Further progress depends on the full stack: infrastructure for faster training and real-time inference, pretraining data at greater scale and coverage, and models that transfer to robots with less adaptation.
We expect it to be a long-term research and engineering effort.
4
22
18,281
Perceptive Humanoid Parkour (PHP) is now open source!
Humanoids can autonomously climb 1.25m walls, vault obstacles, and chain complex parkour skills using onboard depth perception.
Code: github.com/amazon-far/php_pa…
Demo: php-parkour.github.io/demo.h…
Our first podcast episode is finally live!
I personally loved the casual chat after the presentation (From 46:05 onwards).
Sorry about the audio—my mic didn't record, so I had to re-record my lines as best as I could remember them.
And huge thanks to @LujieYang0 and @chris_j_paxton.
Looking forward to seeing more research published on our Booster robots 😎
1
18
1,688
Yi Yang developed a simulation-trained policy that lets robots precisely whip ropes toward any target from a specified direction, achieving 87% real-world positional accuracy.
Whip the pepper!
A robot swings a rope and hits a falling balloon mid-air. Motion capture predicts where it will be; the policy strikes there.
Point It, Strike It: send a rope tip to any 3D point from a chosen direction. Across 25 targets on three real ropes, 87% of strikes land within 5 cm.
1
2
10
1,082
DexWeave: Dexterous humanoid loco-manipulation from human demonstrations.
Naichuan Sun and his team developed a framework that converts human motion data into executable humanoid robot actions, combining locomotion, whole-body coordination, and dexterous hand manipulation.
Using motion retargeting and reinforcement learning, they demonstrated the approach on a Unitree G1 with dexterous hands.
Paper: arxiv.org/abs/2609.34724
Project & Videos: dexweave.github.io/
12
19
185
8,698
FIRED and SUED BY BYTEDANCE. NOW VALUED AT $200 MILLION.
Keyu Tian, A former ByteDance intern who was fired for interfering with AI model training and later sued by the company.
Now he’s building his own world-model startup.
• $30M raised
• $200M post-money valuation
• Backed by 5Y Capital and IDG Capital
• Just ~10 employees
• No publicly announced company name
• No released product
• First full model expected in 2027
And here’s the interesting part: he’s also a first author of a NeurIPS 2024 Best Paper.
From being fired and sued by ByteDance to raising nearly $30M from top-tier VCs.
That’s the AI funding market we’re living in.
5
29
3,706
Congrats @joshavata and the Mecka team!
Data is the bottleneck in physical AI. Robots don't lack intelligence, they lack experience.
Great to see Sequoia backing a team that goes out and captures it, through a Global network of fleets!
Paid partnership (ad)
12
1,906
Congrats Sandor! And hi @lukas_m_ziegler
🚨 BREAKING:
Minerva Humanoids (@mnv_humanoids) has just emerged from stealth with $10M in pre-seed funding led by @generalcatalyst.
It's building humanoid robots for the world's most dangerous jobs. Oil and gas. Explosive ordnance disposal. Hazardous material response. ☢️
Every 75 minutes, an energy worker somewhere in the world is killed on the job.
That's roughly 7,000 deaths a year. NATO consistently identifies EOD and hazmat response teams as the highest-risk segment in every unit, relying on decades-old wheeled, single-arm robots that simply lack the dexterity for complex tasks.
Minerva's answer is Roger, a purpose-built semi-autonomous humanoid that puts the specialist's judgment and hands on the task while keeping their body in a command post far from the danger.
Founded by CEO Sandor Felber and CTO Maurice Rahme, with first paid pilots launching this fall.
The investor lineup is strong with General Catalyst, @LongJourneyVC, @CredoVentures, MVP Ventures, and many other cool funds!
I am proud to be among the angel investors in this round! 😇
Having spent more than 2 years deploying robots in the energy and oil & gas sector, I know firsthand how real this problem is, and how hard it is to solve.
Minerva is going after exactly the right market, with exactly the right approach.
This is the kind of company robotics was always supposed to build! 🛢️
🔗 Read more here: finance.yahoo.com/technology…
~~
♻️ Join the weekly robotics newsletter, and never miss any news → ziegler.substack.com
1
23
1,682
Been knowing them for over a year since stealth. Congrats on the big strap.
Meet Roger, the humanoid purpose-built to protect human life in the most dangerous jobs across energy and public safety.
The experts stay safe and in control. Roger goes in.
No full autonomy needed.
Roger takes the risk. They get the job done.
Idea to walking in 5 months.
1
21
1,743
Been playing with Hedwig a bit. At first glance it reminded me of Meta’s Muse, but after actually using it, I think they’re approaching the agent problem pretty differently.
Muse is more agent-first: you delegate a task and the agent goes off to execute it.
Hedwig is more workspace-first: your apps, context, workflows and agents live together on the same visual canvas, so you can work with and steer the agent in context.
Congrats on the launch.
we’re going to kill a one trillion dollar industry.
we’re building Hedwig: one visual workspace that connects your tools and gives your agent the full picture, and going after the biggest players like google and meta.
built by a team from stanford, meta, and polymarket.
loved by 6,000+ across the world, comment + RT for free access
hedwigmail.com
1
5
720
deformx.github.io/
Credit: Yi Yang, Xiang Fei, Lehong Wang
1
11
5,166
