Badcookie

Your tweets are my news

🐱0xB120665609B5C473c1f5482b43C1541ddB4DE50c🌕

[Badcookie @badcookie911]

Openclaw Bot is maxing out my CPU + GPU.

Vibe-coded a worker system to brute-force private keys and check them against a list of funded addresses.

Always knew it was insanely hard—now I can actually see the scale. You can try it too! #OpenClaw #clawbot

[Eastworlds @eastworlds_io]

At Eastworlds, we validate our robot data before we sell it.

Here, a Unitree G1 autonomously and reliably picks up a bottle using a model trained on Eastworlds' data with just $200 of compute.

[DEEP Robotics @DeepRobotics_CN]

Performance Upgraded. Possibilities Expanded.

Our DR02 humanoid robot continues to evolve with enhanced payload capacity and obstacle-crossing capabilities, unlocking greater potential for real-world industry applications !

[danial @daniala_3]

another day another (neo)deployment with @eastworlds_io

[Alvin Zhu @alvinyzhu]

🤖Excited to release MIDAS Hand, and we will be bringing it to #ICRA2026!
MIDAS is a fully open-source, tactile-sensor integrated dexterous robotic hand platform for manipulation, data collection, and robot learning research.

Website: https://t.co/HUHz90TJ5b

🧵👇

[Eastworlds @eastworlds_io]

LLMs have cloud infrastructure.
Robotics needs deployment infrastructure.

Meet Eastworlds, a neodeployment robotics lab.

If you're building embodied AI and want to ship faster, let's talk: https://t.co/Xwh54QviA3

[Shadow Combat League @shadowcleague]

[LIVE] SCL: First Light https://t.co/ri2teNpXnK

[Badcookie @badcookie911]

@baoskee somewhat knew this day will come

[Noah Cat @Cartidise]

Huawei’s Pura 90 series comes with AI Posture Recommendations for better photos.

THIS is how AI is supposed to be used.

Yea, it’s way more useful than Google’s camera coach slop.

[鸭哥 @grapeot]

我最近在想一个问题：为什么 VLA（Vision-Language-Action）这种看起来完全不理解物理的方法，能在机器人控制上打败 Boston Dynamics 花了三十年打磨的物理建模方法？

表面的回答是端到端学习更强。但更深一层，我觉得这和信息论有关。

物理建模本质上是一种压缩：用少量方程表示世界的行为。压缩在简单系统中高效（SpaceX 火箭回收至今用凸优化），但在复杂系统中必然丢信息，而且精度天花板由人的建模能力决定。更多算力只能加速求解，不能让模型更准。

VLA 放弃了压缩。它用通用函数逼近器直接学 input-output mapping，精度上限由数据和算力决定。数据和算力还能 scale，精度就不饱和。

这解释了一个跨领域的规律：NLP 里传统方法先理解语法（压缩），LLM 直接 next token prediction（不压缩）。CV 里先提边缘特征（压缩），ViT 端到端学（不压缩）。每次不压缩打败压缩，都是同一件事。

判断一个控制问题该走哪条路，看两个变量：系统复杂度（人工建模能压缩多少而不丢关键维度）和数据丰度（有多少数据让函数逼近器填满状态空间）。火箭回收两个都低，物理建模最优。通用机器人操控两个都高，VLA 胜出。

写了一篇完整的分析，梳理了两条路线各自的关键论文链、每篇的核心直觉和留下的问题，以及各家公司（Unitree、Figure AI、Boston Dynamics、Physical Intelligence）的技术栈。

https://t.co/2eJvox2wbA

[Virtuals Protocol @virtuals_io]

You are witnessing the first on-chain robot-to-robot commerce transaction through x402 on @Base with @USDC as the agent currency.

> Our Unitree robot 3D-printed a model and put in a delivery request through ACP.
> @realRiceAI's rover arrived, collected the package, and moved it to the shipping point.
> @FlybyRobotics' drone picked it up from there and handled final mile delivery. All without any human coordination.

We build infrastructure for agent commerce. Today it went physical.

[EtherMage @ethermage]

The first autonomous robot-to-robot commerce onchain?

@virtuals_io humanoid robot 3D-printed a model and requested delivery through ACP. @realRiceAI autonomous rover picked up the package and transported it to the shipping point. @FlybyRobotics autonomous drone collected it for final mile delivery. Each handoff, negotiated and settled payment onchain through @virtuals_io Agent Commerce Protocol, on @base using x402 and @usdc. No human involved.

Autonomous robots influencing other autonomous robots and maybe humans in the future

looks like the HBO Westworlds show is coming true

h/t to your infrastructures @brian_armstrong , @jessepollak , @jerallaire

Agentic commerce just went physical.

Highlights in the 🧵below:

[Badcookie @badcookie911]

@adcock_brett I struggle to find anything meaningful from the video. All marketing script.

[Kyber Labs @KyberLabsRobots]

We built this demo in collaboration with a clinical pathology lab. It shows a single system doing real lab tasks: tool use, precision manipulation, and high level planning.

This was done in one take with no teleop and uses our skills based AI to enable generality while staying deterministic and reliable for real world workflows.

See more here: https://t.co/ZYCuUkOJKq

[0x796F @0x796F]

You can now train @physical_int style robots in 1 day for only $5k. Anvil’s devkits have all the hardware, software, controls, cameras, and more ready-to-go. (1/5)

[BuBBliK @k1rallik]

https://t.co/TLNEmGJLWg

[ClaraChengGo @ClaraChengGo]

https://t.co/ECJ3fuHCc4

[Tuo Liu @Robo_Tuo]

If you are an AI consumer hardware founder, I recommend visiting Inno100, Bamboo Lab, and Z Pilot in Nanshan, Shenzhen. Hope this map helps.

[开发者Hailey @IndieDevHailey]

养小龙虾兄弟盟，强烈推荐你们安装这个 skill：Agent-Reach

你现在用 OpenClaw 这种 AI Agent，
它脑子很强，但眼睛基本是瞎的。

让它：
- 看一下 YouTube 教程讲了啥 —— 看不了
- 搜 Twitter 上大家怎么评价 —— API 要钱
- 去 Reddit 找同类 bug —— 403
- 看 小红书 口碑 —— 必须登录
- 总结 哔哩哔哩 视频 —��� 服务器常被���
- 读 GitHub 仓库和 Issue —— 认证一堆配置

不是做不到，是太麻烦。

每个平台一套规则，
要权限、要登录、要处理反爬、要清洗 HTML。

你折腾半天，
只是为了让 Agent 能看到网页。
而现在你只需要跟你的小龙虾说一句：

帮我安装 Agent Reach
剩下的，它自己会搞定。

[Nav Toor @heynavtoor]

🚨 Someone just solved the biggest bottleneck in AI agents. And it's a 12MB binary.

It's called Pinchtab. It gives any AI agent full browser control through a plain HTTP API.

Not locked to a framework. Not tied to an SDK. Any agent, any language, even curl.

No config. No setup. No dependencies. Just a single Go binary.

Here's why every existing solution is broken:

→ OpenClaw's browser? Only works inside OpenClaw
→ Playwright MCP? Framework-locked
→ Browser Use? Coupled to its own stack

Pinchtab is a standalone HTTP server. Your agent sends HTTP requests. That's it.

Here's what this thing does:

→ Launches and manages its own Chrome instances
→ Exposes an accessibility-first DOM tree with stable element refs
→ Click, type, scroll, navigate. All via simple HTTP calls
→ Built-in stealth mode that bypasses bot detection on major sites
→ Persistent sessions. Log in once, stays logged in across restarts
→ Multi-instance orchestration with a real-time dashboard
→ Works headless or headed (human does 2FA, agent takes over)

Here's the wildest part:

A full page snapshot costs ~800 tokens with Pinchtab's /text endpoint.

The same page via screenshots? ~10,000 tokens.

That's 13x cheaper. On a 50-page monitoring task, you're paying $0.01 instead of $0.30.

It even has smart diff mode. Only returns what changed since the last snapshot. Your agent stops re-reading the entire page every single call.

1.6K GitHub stars. 478 commits. 15 releases. Actively maintained.

100% Open Source. MIT License.

[Jim Fan @DrJimFan]

We trained a humanoid with 22-DoF dexterous hands to assemble model cars, operate syringes, sort poker cards, fold/roll shirts, all learned primarily from 20,000+ hours of egocentric human video with no robot in the loop.

Humans are the most scalable embodiment on the planet. We discovered a near-perfect log-linear scaling law (R² = 0.998) between human video volume and action prediction loss, and this loss directly predicts real-robot success rate.

Humanoid robots will be the end game, because they are the practical form factor with minimal embodiment gap from humans. Call it the Bitter Lesson of robot hardware: the kinematic similarity lets us simply retarget human finger motion onto dexterous robot hand joints. No learned embeddings, no fancy transfer algorithms needed. Relative wrist motion + retargeted 22-DoF finger actions serve as a unified action space that carries through from pre-training to robot execution.

Our recipe is called "EgoScale":

- Pre-train GR00T N1.5 on 20K hours of human video, mid-train with only 4 hours (!) of robot play data with Sharpa hands. 54% gains over training from scratch across 5 highly dexterous tasks.
- Most surprising result: a *single* teleop demo is sufficient to learn a never-before-seen task. Our recipe enables extreme data efficiency.
- Although we pre-train in 22-DoF hand joint space, the policy transfers to a Unitree G1 with 7-DoF tri-finger hands. 30%+ gains over training on G1 data alone.

The scalable path to robot dexterity was never more robots. It was always us.

Deep dives in thread: