The cool Boston Dynamics results from yesterday are based on Toyota large behavior models, which are not large pretrained VLM foundation models. In the end the bottleneck for robotics is... robotics. Good data good hardware etc. Don't be disempowered.

It seems like there is not enough of a policy response to the fact that, with 57M miles of data, Waymo’s autonomous vehicles experience 85% less serious injuries & 79% less injuries overall than cars with human drivers. 2.4 million are injured & 40k killed in US accidents a year

Very honoured to have our work on generalized legged locomotion published in Science Robotics Science Robotics ! (with Junzhe (JJ) He Robotic Systems Lab ) (I show up on the left of screenshot here) Here is a thread of takeaways and some personal opinions.

Funny. HTX in Chinese tech community is seen as one having bad tech skills but strong media skills to become a tech influential. If you call any professional technician "you are like HTX" it's kinda an offense. But this won't deny HTX from the popularity for common audience.

I'm super excited to announce mjlab today! mjlab = Isaac Lab's APIs + best-in-class MuJoCo physics + massively parallel GPU acceleration Built directly on MuJoCo Warp with the abstractions you love.

Wait for it

Very clean and applied work. The next 3 years are going to be crazy for humanoid demos.

Implementing motion imitation methods involves lots of nuisances. Not many codebases get all the details right. So, we're excited to release MimicKit! github.com/xbpeng/MimicKit A framework with high quality implementations of our methods: DeepMimic, AMP, ASE, ADD, and more to come!

I've long wondered if we can make a humanoid robot do a 𝘄𝗮𝗹𝗹𝗳𝗹𝗶𝗽 - and we just made it happen by leveraging 𝗢𝗺𝗻𝗶𝗥𝗲𝘁𝗮𝗿𝗴𝗲𝘁 with BeyondMimic tracking! This came after our original OmniRetarget experiments, with only minor tweaks to RL training: relaxing a

Works like this deserve a lot more attention. In embodied AI, the most common argument against physics-based simulation is that deformables are hard/expensive to simulate. Offset Geometric Contact just made it 300x faster. It computes vertex-specific displacement bounds to

I swear a year ago these things would only shuffle along in a straight line and fall over if you looked at them wrong

Unitree G1 crawl policy deployed to hardware! Plenty of room for improvement, but it's a start.

Real-world deployment is crucial for robotics. I wrote about a recipe: strong base model + RL post-training (offline & online) → toward human-level capability. lei-kun.github.io/blogs/RL100.ht…

Didn’t go max speed for safety and space reasons but pretty happy with the result!

Smokey says: "Only YOU can prevent boring costumes." We're out patrolling the robot park this Halloween - keeping our circuits safe and our actuators calibrated. Have a happy and safe Halloween from all of us at Apptronik! 🤖 🐻 🌲 🎃 #apptronik #ai #humanoid #humanoidrobot

So, AI video -> GVHMR -> sim2real motion tracking works, btw This is a very tricky and dynamic motion that was trained on a video I generated on fal using Google Veo 3.1 Fast You can literally do prompt -> robot motion now, this is crazy lol

K-scale has shut down. Huge kudos to the team and the amazing work they've done over the last year & their belief in open-source first. We had fun playing with the very first K-Bot at Dimensional HQ. It's clear the US manufacturing base is far from being able to support

モーションキャプチャシステムを使った人型ロボットの動作訓練 滑らかで人間のようなダンスを実行する youtu.be/6TcwjKAQTIo #humanoidrobot #MotionCapture #RealTime #raining #dancing #ReinforcementLearning #ImitationLearning #VDSuitFull #Virdyn