Pinned Tweet
Check out our study in @NatMachIntell, where we use AI copilots for brain-computer interfaces (BCIs): nature.com/articles/s42256-0…. We enable a paralyzed participant to neurally control a computer cursor and a robotic arm (non-invasive, no surgery). Video here: youtube.com/watch?v=QSEKLGOQ…
Jonathan Kao retweeted
We introduce Token Steering: an inference-time method for steering pretrained autoregressive VLAs through direct intervention in their action-token space—no retraining or fine-tuning required.
Excited to share flow control, a method that allows users to steer VLAs in real time with keyboard arrow presses. This method doesn't require any fine-tuning of a VLA and can run out-of-the-box.
Flow control sets the initial condition of the flow matching action expert to be the user's action. Keyboard inputs are crude, but the flow matching action expert then transforms the action to be a high quality on-policy action that reflects user intent.
One reason we're excited about this is because we believe future assistive robotics, like in brain-computer interfaces, can involve users steering robot policies with low-dimensional, crude, and noisy inputs. Read more about the paper here: jcykao.github.io/flowcontrol…
Jonathan Kao retweeted
Excited to introduce LightBeam, a CTC decoder for speech neuroprostheses that drastically cuts memory load while achieving state-of-the art (SOTA) results
Paper: arxiv.org/abs/2603.14002
Code: github.com/ebrahimfeghhi/bra…
Co-authors: @hu_is_lionel @nrhadidi @JonathanCKao
Thread 🧵
Jonathan Kao retweeted
We are excited to share our latest work in @ScienceMagazine studying cooperative behavior in both biological and artificial intelligence systems — fantastic work by @JanineJiang22, Linfan Gu, and wonderful collaboration with @JonathanCKao! science.org/doi/10.1126/scie…
Jonathan Kao retweeted
Thrilled to share that our paper “Neural Basis of Cooperative Behavior in Biological and Artificial Intelligence Systems” is now out in @ScienceMagazine. Amazing collaboration with Linfan Gu - huge thanks to @TheHongLab and @JonathanCKao for their guidance. grateful for the
Thanks to Johannes Lee, Sangjoon Lee, Abhishek Mishra, Xu Yan, Brandon McMahan, and the rest of our team for their awesome work on this study! Thanks also to Nature News for covering our paper! nature.com/articles/d41586-0…
We believe shared autonomy will play an important role in future BCIs (both invasive and non-invasive)! We're hiring postdocs, particularly in robotics + AI. If you're interested in joining our team, please visit: kaolab.org/join.
Jonathan Kao retweeted
We develop a speech neuroprosthesis that is accurate and can be efficiently adapted at test-time — bringing us closer to real-time communication on portable devices for patients with severe paralysis.
Paper: arxiv.org/abs/2507.02800
Thread ↓
Excited to share this study, a collaboration with @TheHongLab, where we identify shared neural dynamics in rodents and AI during social interaction! In AIs, we also perturb these dynamics and show they are causal to social behavior. Awesome work by @xzh_brain and Nguyen Phi!
Thrilled to share our latest work in @Nature studying inter-brain neural dynamics in both biological and artificial intelligence systems — amazing work by @xzh_brain, Nguyen Phi, and fantastic collaboration with @JonathanCKao! Read the full article here: rdcu.be/euu5Z
We distill the benefits of hierarchical RL into a simpler flat policy for offline goal-conditioned RL, achieving SOTA performance on very long-horizon, high-dimensional control problems.
See below for more!
Hierarchical methods for offline goal-conditioned RL (GCRL) can scale to very distant goals that stymie flat (non-hierarchical) policies — but are they really necessary?
Paper: arxiv.org/abs/2505.14975
Project page: johnlyzhou.github.io/saw/
Code: github.com/johnlyzhou/saw
Thread ↓
1/4 Check out our preprint, “Non-invasive brain-machine interface control with artificial intelligence copilots.” (biorxiv.org/content/10.1101/…). We demonstrate a paralyzed participant has non-invasive neural control of a cursor and a robotic arm (video: youtube.com/watch?v=SFQDkMmc…).
3/4 We believe AI copilots can help improve brain-machine interfaces significantly and we’re doing a lot of work in this direction. Relatedly, check out our new shared autonomy approach, IDA (arxiv.org/abs/2409.15317), recently accepted at NeurIPS (video: youtube.com/watch?v=luqypU-H…).
4/4 This work was a big team effort led by Johannes Lee, @abhishek_iitkgp, Sangjoon Lee, @XuKristenYan, and Brandon McMahan, all trainees at UCLA!
Jonathan Kao retweeted
What do LLMs map to in the brain? In some datasets, not much. We emphasize the need for simple controls when analyzing the neural predictivity of trained and untrained LLMs.
arxiv.org/abs/2406.01538
In collaboration w/ @ebrahim_feghhi, supervised by @IbanDlank and @JonathanCKao
Jonathan Kao retweeted
ND -> 1st %! Poured my heart into my 1st NRSA submission and was absolutely gutted when it wasn’t discussed. Extremely thankful for my advisors @SuthanaLab @JonathanCKao for dusting me off and teaching me how to make others just as excited about my ideas as I am. Bag secured🫴💪
