1/ Introducing SN/08: RoboPlayground on TeleArms.
@BitRobotNetwork111 тыс. подписчиков
75 000 просмотров
37 комментариев
Комментарии из оригинального поста в X (Twitter)

@BitRobotNetwork
2/ Robotics has a data problem. The best training data comes from teleoperation — but it doesn't scale. You need a lab, hardware, people. Simulation helps, but sim data lacks the 'human touch' that makes it useful for policy training. SN/08 solves this: combining both worlds with teleoperate-in-sim to remove the hardware bottleneck.
15 отметок «Нравится»

@BitRobotNetwork
3/ SN/08 builds on the RoboPlayground paper from @UW and @allen_ai researchers: When a manipulation task is described in natural language → the system spins up dynamic eval tasks in sim. Testing policies this way reveals failures that fixed benchmarks miss. More contributors = more diversity in evaluation. Now pair that with thousands of teleop-in-sim hours from the BitRobot community…
11 отметок «Нравится»

@BitRobotNetwork
4/ We're bringing RoboPlayground tasks into SN/08. Lift Pot. Rotate Valve. 30+ tasks now, 100+ more coming. RoboPlayground spins up rich evaluation challenges. BitRobot turns them into teleop-in-sim games — generating 2 things: - high-quality human demos for imitation learning - evaluation data that exposes real policy failures
7 отметок «Нравится»

@BitRobotNetwork
5/ Getting started is simple: choose your keyboard, Joy-Con motion controller, or other supported device. Tasks range from easy to challenging. As you play, you'll earn ELO ratings that rank your skill across task types and levels — and against other players.
8 отметок «Нравится»

@BitRobotNetwork
6/ SN/06 + SN/07 capture high-fidelity teleop data from real hardware (LeKiwi and XLeRobot). SN/08 adds the simulation layer to the stack. All 3 subnets share the same mission: creating the most comprehensive, diverse open dataset of how humans manipulate the world.
7 отметок «Нравится»

@BitRobotNetwork
7/ Thrilled to build on the foundational work by @DJiafei, @yiruhelenwang, and the RoboEval + RoboPlayground team ( @carter_t_ung, @siddhss5, @fox_dieter17849, Evan Gubarev, Christopher Tan, and others). RoboPlayground brings dynamic eval tasks into sim — BitRobot turns them into teleop games for a global contributor base, yielding richer, more diverse datasets that help close the sim-to-real gap.
9 отметок «Нравится»

@BitRobotNetwork
8/ Early @BitRobotNetwork members are already testing SN/08. We've reached over 27K+ completed game missions w/ 400+ hours of valuable training data collected Join the TeleArms waitlist for a chance to contribute and earn rewards:
13 отметок «Нравится»

@Tola_niii
the cube handover and lifting the pot are my favourites 🙂 Im currently ranked 5th on the RoboPlayground leaderboard 🔥
4 отметки «Нравится»

@targetraj2
Love this approach making robotics training accessible through simulation is a smart way to scale real world innovation.
3 отметки «Нравится»

@bibsvault
feels good to be part of this
2 отметки «Нравится»

@nonibond
Keyboarding is hard work. Double points please
2 отметки «Нравится»

@shelleymaeph
congratulations! I need to up my game hereeee
1 отметка «Нравится»

@phuong1492
AI has long lacked high-quality data, while games naturally maintain strong engagement. BitRobot is using gaming as a “cheat code” to solve the data problem for embodied AI. When users genuinely want to play, they unintentionally generate more natural and valuable datasets compared to traditional forced training approaches.
1 отметка «Нравится»

@jamyyip
BitRobot Network is the ultimate master key to the Embodied AI era , powering the rise of Humanoid Robotics through a robust DePIN infrastructure, attracting smart money from industry giants, and building a high-potential Skill App Store for savvy hunters to capitalize on
1 отметка «Нравится»

@Bearcrypto2021
That's a perfect update. LFG 😘😘😘
1 отметка «Нравится»

@bachkhoabnb
LFG 🫡🫡
1 отметка «Нравится»

@Habibiofcrypto1
This is definitely something we need
1 отметка «Нравится»

@phamminh5959
Năm nay AI sẽ dẫn đầu xu hướng mới

@NangiETH
Controlling robots with a keyboard is great, and training real AI is even better. I hope you guys continue to build on these positive things.

@amaltvm2013
I like that this goes beyond typical AI narratives and focuses more on connecting intelligence with real-world systems. As AI starts interacting directly with environments, this kind of infrastructure could become increasingly important.















