humanoid-robot —— 最强机器人大脑 SkildBrain

西京机器人短讯178期
任何一台机器人上
最强机器人大脑 SkildBrain
@jingxisaibo
这只机器人，装了全球最强，机器人大脑 #人工智能机器
人 #机器人 humanoidrobot #robot

资料地址：

https://www.youtube.com/shorts/mhZEGU1fzSQ

亮点:

运控模式自动切换或者集成为同一个模型。

Unlike prior robotics models limited to scripted movement or narrow tasks, Skild Brain learns through both large-scale simulation and real-world robot data, addressing longstanding data shortage challenges in the field.

Key Demonstrated Abilities:

Climbing stairs and maintaining balance after being nudged
Picking up objects in cluttered, unpredictable settings
Performing household-style tasks such as sorting dishes or operating in kitchen-like environments.
These demonstrations highlight the combination of spatial reasoning, adaptability, and safety mechanisms—including built‑in power limits to prevent excessive force around humans.

Skild Brain is unique for its omni-bodied intelligence: a single model that works across robot morphologies and applications. This contrasts with typical AI approaches that train one model per robot type.

The data puzzle

One of the biggest challenges in building a robotics foundation model is the lack of any large-scale robotics data, and collecting real-world data using hardware is slow and prohibitively expensive. Skild AI has tackled this challenge head-on: achieving scale by leveraging large-scale simulation and human videos on the internet to pre-train its foundation model at scale. Skild AI post-trains this foundation model with targeted real-world data to deliver working solutions to its customers.

Skild trains models that can work across different morphologies, including human data, vastly expanding the available training set. This has enabled Skild to build a truly physical foundation model, which can work on quadrupeds, humanoids, tabletop, and mobile manipulators. Trained on diverse data, tasks, and hardware, Skild’s general-purpose brain is fine-tuned and distilled to meet deployment needs.

The team

Skild AI’s team has been behind key robotics breakthroughs such as self-supervised learning, curiosity-driven exploration, large-scale imitation, sim2real locomotion, parkour, dexterous manipulation, learning from internet-scale video, and more. Team members have also built production-ready systems at Anduril, Tesla, Nvidia, Meta, Kitty Hawk, Google, Everyday Robotics, and Amazon.

Skild AI is backed by investors including Lightspeed Venture Partners, Coatue, SoftBank Group, Jeff Bezos (through Bezos Expeditions), Felicis Ventures, Sequoia, Menlo Ventures, General Catalyst, CRV, Amazon, SV Angel, and Carnegie Mellon University.