Bo Dai

@daibond_alpha

Assistant Professor at Georgia Tech @GTCSE, Senior Staff Research Scientist at @GoogleDeepMind | ex @GoogleBrain

California, USA

bo-dai.github.io

Joined October 2012

830 Following

3,050 Followers

Replies Reposts Media

Pinned
Bo Dai
@daibond_alpha
Oct 15
One Kinex for any robot, making scaling across different robots possible!
Chenxiao Gao
@chenxiaogao
Oct 15
Introducing Kinex, the harness that connects coding agents to any physical robot, unlocking scalable learning in the physical world. Tell Kinex about your robot; it then carries out tasks and learns from every run. We demonstrate Kinex across diverse real-world applications in a
2
Bo Dai
@daibond_alpha
1h
Finally, finally. We have delivered spectral representation as skills for real robot! Please check this astonishing result and more results along this line are on the way!
Feiyang Wu
@fei_yang_wu
6h
Humans learn how to solve tasks by composing what they have learned. Can robots do the same? Can robots walk while squatting once they learn individual motions? Introducing Spectral Skills. Steer, mix, and chain motion skills through composing energy functions.🧵
1
Bo Dai
@daibond_alpha
Oct 29
Gimini 4 is capable of robotics engineering
rlebench
@rlebench
Oct 29
🚀 Gemini 4 Argon takes #1 on 4 robotics tasks in RLE-Bench! The latest Gemini 4 Argon results are in: 72.4 RLE Index, ranking #2 overall, just behind GPT-6 Astra (73.4). More impressively, Gemini 4 Argon takes the top spot on 4 out of 9 tasks: * Harness Engineering * Embodied
Bo Dai
@daibond_alpha
Oct 27
With our new self-improving agent architecture, we equipped our kinex physical agent with RSI ability and demonstrated its surprising efficiency for solving *all* tasks in RoboCasa with only a handful experiences Kinex: kinex-agent.org Physical RSI:
Rushi Qiang
@rqiang_rushi
Oct 27
Coding agents can solve all RoboCasa tasks from scratch, by self-evolving its skills and tools from its own trials. Our system grew its library from empty to 6 canonical skills and 70 functions in seven rounds, and solved all 365 tasks. See details: stacx-lab.github.io/physical-rsi/
Bo Dai
@daibond_alpha
Sep 23
It is very similar to our instruction following embedding model used for instruction aware retrieval arxiv.org/pdf/2505.21439
Matija Sosic
@MatijaSosic
Sep 16
Here's a 45-second TL;DR on Jev. I find the core idea beautifully simple, but the video made it really hard to understand. Hope you find it helpful.
1

Bo Dai

Bo Dai’s posts