곰엔 retweeted
新ランドセル、収縮で微妙にネジ穴がずれたりしてめちゃ大変だったけどとりあえず取り付け!
バッテリーも既に心許なかったので容量多めのタイプに変更
ランドセル バッテリー新構成でまずはCM5無しの従来仕様で火入れ、ヨシ！
#Sim2Real #ホビロボ制御
1
2
22
1,056
In parallel i am going to build the visual cortex of these zombies while validating Sim2Real of the spinal cord. I am pursuing multiple additional projects which will all come together. I see a narrow way through to achieve the PINNACLE OF AERIAL SUPERINTELLIGENCE!! JUST WATCH.
21
I have successfully achieved a foundation drone policy that can adapt to any standard existing off the shelf drone and control it. Sim2Sim validated, next step is Sim2Real and for that i need to make some modifications to my FC and RC transmitter. 1/
1
34
Li Feifei's team, with NVIDIA, presents SimFoundry: a system that breaks the Sim2Real gap. It auto-generates interactive, evolvable sim environments from one real video, slashing costs and reducing model transfer performance loss, enabling a new Real2Sim path for embodied AI training.....
李飞飞团队联合英伟达等机构推出SimFoundry系统，突破Sim2Real困境。它仅凭一段真实世界视频，即可自动生成可交互、可进化的仿真环境，大幅降低搭建成本，缓解模型迁移的性能折损，为具身智能训练开辟Real2Sim新路径。
41
周观点：物理AI催化密集，网易入通，关注快手AI资产重估
[红包]传媒：本周物理AI产业催化密集，重点关注 ￥风语筑。三星加码物理AI产业链，宇树科技科创板IPO获注册，产业趋势加速验证。风语筑与松应科技设立风应科技，聚焦物理仿真引擎、Sim2Real训练及空间合成数据，有望从传统数字展示工程商升级为“空间数据 仿真训练 智能交互”平台，同时应收账款化债改善现金流，成长与主业修复共振。
[太阳]游戏：￥网易 完成双重主要上市转换并正式纳入港股通，重点看好稳健基本盘 新品周期 自研出海 高股东回报。暑期旺季开启，游戏基本面迎关键拐点，关注网易《遗忘之海》、腾讯《失控进化》、巨人《超自然》、腾讯《洛克王国》、心动《小镇》等产品弹性。
[太阳]云模芯：
[玫瑰]Meta拟对外出售富余AI算力，布局AI云，我们认为有助于缓解AI Capex ROI担忧，提升资源利用效率，也强化互联网CSP“基础设施 生态入口”逻辑。海外AI算力不是需求见顶，而是进入更重视现金流与ROI的健康扩张阶段。
[玫瑰]可灵完成大额融资，关注 ￥快手 AI资产重估。可灵本轮融资约28亿美元，后续上限约30亿美元，投后估值约180亿美元，阿里、腾讯、百度等产业资本参与；当前ARR预计超5亿美元，AI视频有望带动快手内容生态与估值重估。
[玫瑰]模型：￥智谱、￥MiniMax 下周迎来解禁，短期或有交易波动，但不改模型资产稀缺性。我们继续看好国产大模型在Coding、Agent、长上下文和性价比上的迭代，关注后续融资、模型升级及商业化进展。
[太阳]新消费：￥布鲁可 新品类积木人表现出色，出海依托迪士尼等IP扩充预计高增；￥泡泡玛特 继续关注IP产品周期、新品类、海外运营优化及渠道调整。
369
HomeMadeGarbage retweeted
私とSim2Real
x.com/H0meMadeGarbage/status…
x.com/H0meMadeGarbage/status…
x.com/H0meMadeGarbage/status…
HomeMadeGarbage
2
21
1,523
17h
DexForce raised RMB 1 billion (approximately US$139 million) in Series B funding at a post-money valuation above RMB 10 billion (approximately US$1.4 billion) this week.
Founded in 2021, the Shenzhen company built its core technology around DexVerse, a Sim2Real embodied-intelligence engine that trains robots in simulation before real-world deployment.
It now operates as both a software platform and a hardware manufacturer, shipping the W1 Pro humanoid alongside its simulation stack.
H1 2026 revenue approached RMB 100 million (approximately US$14 million), with full-year targets of RMB 250 million to 300 million (approximately US$35 million to 42 million) and nearly 1,000 humanoid shipments planned for 2026.
The round drew state-linked and industrial capital: Shenzhen Capital Group, Qianhai Mother Fund, ICBC Capital, Hengjian Assets, Guiyang Digital Economy Fund, and Lens Technology.
Source: stcn.com/article/detail/3988…
Profile: korthos.xyz/ecosystem/compan…
4
229
everett Damasco retweeted
AI Sapiens K1’s Sim2Real package connects robot feedback, operator/API input, policy inference, and joint command publishing into a 1000 Hz control loop for real-robot deployment.
Built-in modes include Damping, ReadyPose, Velocity, Mimic, and more, with failsafe logic checking RC input, IMU, joint states, policy actions, and body orientation before execution.
Train. Validate. Deploy on hardware.
More AI Sapiens K1 development info coming soon.
Docs:
docs.robotis.com/docs/system…
#ROBOTIS #AISapiens #Sim2Real #PhysicalAI #HumanoidRobot #Robotics #ROS2 #DYNAMIXEL #RobotLearning
2
16
74
3,899
Jul 5
Its the big question of sim2real, we saw different examples like SimToolReal and other dexterous manipulation methods successfully deployed in real robots, but we still think there is some gap, real depends on the tasks and the level of complexity ( real envs in sim cost a lot, another factor)
1
16
HomeMadeGarbage retweeted
Jun 30
PWM micro servo biped robot
— Sim2Real with AI-optimized motion —
1
2
33
3,632
strnam retweeted
AI/ML is amazing saaarr!
Physical AI solves it saarr!
JEPA does wonders saarr!
We only care about Sim in Sim2real saarr!
Jul 3
Lots of people are entering robotics by skipping the foundations.
The pitch is that world models understand physics so you don't need them.
Robot physics was solved decades ago. Calibration, motion planning, collision avoidance. Known quantities, not open research.
I get the appeal. Train one model, let it handle calibration, motion, collisions, and the task itself.
But watch the demos these teams ship. Autonomous footage played back at 4-10x. That's not a hardware limit. That's a team without foundations.
Strong foundations allow you to show your videos at 1x proudly.
3
4
48
11,619
This what happens when you do too much sim2real instead of model predictive control.
1
14
3/ the problem with learning a dynamics model in sim like Isaac simulation is you will always have the sim2real mismatch. You should be training a transformer to learn to simulate the cloth movement instead and using that in a low latency loop for a solver to move the cloth.
1
22