Peng Xu
@sippeyxp
Followers
503
Following
36
Media
39
Statuses
193
Working on #Gemini Robotics model in @GoogleDeepMind; lead Gemini Robotics Trusted Tester program; making AI-powered robotics accessible to all.
Near a robot
Joined February 2010
🔥Gemini Robotics On-Device is here! VLA with similar generalization, instruction following, and fast adaptation as our March release, now fits on a 4090!. More exciting: we're 🚀an SDK and a model dev service (flywheel) alongside it 🎯= democratizing model development!.
3
7
28
The robot runs so naturally, yet it is a bit drunk. Someone fed it Si2H5OH?.
Casually spotted: whose robot kid is running around the neighborhood — and bumping into trees?
1
0
6
Both people came and we had a lot of fun yesterday! Saw a lot of innovative and tough requests. People are not shy of placing their personal item on the desk, e.g. two badges on the table and ask who is smarter 🤣 #Gemini model safety get it and avoided an embarrassing moment!.
We took a robot to RSS in LA running our new Gemini Robotics On-Device VLA model. People interacted with the model with new objects and instructions in a brand new environment and the results were amazing!
0
0
7
First (?) VLA demo done outdoor in a conference! The demo is still on! Had a lot of fun. People are very creative in giving instructions! Do not miss it if you are at #RSS2025.
Come by the @GoogleDeepMind booth at @RoboticsSciSys conference in LA!. We’re demoing Gemini Robotics On-Device live, come check it out
0
5
49
Foundation models revolutionize the field of robotics. Something people do as an entire course project before is just a prompt away with Gemini 2.5 Pro. Check it out, find use cases we do not know and show it to us #GeminiRobotics.
We also are sharing a dev blog on how to use Gemini 2.5 for Embodied Intelligence. Try it out on your robot!.
0
0
3
Thanks to a great team of people who made this possible: @xjygr08, @MichaelElabd, @ayzwah, @shahdhruv_, @debidatta, Ken Caluwaerts, Sharath Maddineni, Yuheng Kuang, etc.
0
0
2
The SDK makes using the on-device model easy. Once you have a robot platform (aloha in this case) set up. It should only take a few mins to run the model on the robot.
1
1
3
Thanks for this contribution! The community needs a place for a fair and trusted comparison of policies.
We’re releasing the RoboArena today!🤖🦾. Fair & scalable evaluation is a major bottleneck for research on generalist policies. We’re hoping that RoboArena can help!. We provide data, model code & sim evals for debugging! Submit your policies today and join the leaderboard! :).🧵
0
0
2
At RSS 2025 with a junior intern just in case robotics is not solved by then. Happy to chat!
1
4
229
A great examples from @EnchantedTools, demonstrating Gemini and Gemini Robotics models in human centered robotics applications. Thanks @EnchantedTools, very nice working withing you!.
🚀 Mirokaï x Gemini: A New Era for Embodied AI . Our latest video showing how the Mirokaï are evolving with Gemini 2.0 from @GoogleDeepMind 🤖✨. 🧠 Emotional awareness. 💬 Natural voice-to-voice interaction. 🗺️ Smart 3D navigation. 🔍 Real-world object manipulation. ➡️ Watch how
0
0
5
Want to work with top notch researchers and the active communities and companies, train models build infra to push boundary of embodied AI. Join us!.
Gemini Robotics featured in I/O Keynote by Demis. Last time my work getting I/O Keynote attention is 9 years ago (2016). It was the motion to photon latency reduction (20+ms->2ms) for VR mode and Daydream Ready. Interestingly, in the same I/O Keynote, Sundar talked about a 20%
0
0
7
Gemini Robotics featured in I/O Keynote by Demis. Last time my work getting I/O Keynote attention is 9 years ago (2016). It was the motion to photon latency reduction (20+ms->2ms) for VR mode and Daydream Ready. Interestingly, in the same I/O Keynote, Sundar talked about a 20%
Excited that our work on Gemini Robotics and Gemini spatial understanding have just been featured on #GoogleIO stage! I believe that a frontier model possessing strong real-world understanding capabilities represents the ultimate path to embodied AGI, and we are making rapid
0
0
11
Great work for reaching 2.2 9s. Tracking mastery (near 100% in real scenarios) is a signal for being "serious" (thanks @SudeepDasari) in embodied AI application. Companies should start report 9s or -log10(1-success), ie 1 nine is 90%, 1.3 is 95%. .
Today, we’re thrilled to turn the page on embodied-AI manipulation with a major technical breakthrough that finally delivers PERFORMANCE in real-world conditions. Meet Dynamism v1 (DYNA-1). Some may ask: I’ve seen better robot videos, what’s the big deal? . The difference is.
0
1
8
When did @sundarpichai become funny like this?. Wait! Today is April 14 not April 1st!.
Introducing DolphinGemma, an LLM fine-tuned on many years of dolphin sound data 🐬 to help advance scientific discovery. We collaborated with @dolphinproject to train a model that learns vocal patterns to predict what sound they might make next. It’s small enough (~400M params).
0
0
2
AI is the productivity elixir these days. People who use it well have a python over the assembly level of gains. Wondering when AI will be THE productivity itself, company runs completely by AI agents. 2 years? 5 years?. How about company that fixes your drain pipe?.
I heard this internal memo of mine is being leaked right now, so here it is:
1
0
2
What is the area in which humanoids have the biggest advantage over other types of robots?. Entertainment.
Rock'em Sock'em Robots, but with Unitree G1!. “Unitree will be livestreaming robot combat in about a month — stay tuned! The robot is actively learning even more new skills. You can control it via motion capture, a controller, or voice commands.”
7
2
24