Bryce Grant

PhD @ CWRU | NSF GRFP | Learning what makes robots tick

[Bryce Grant @bryceagrant1]

Your VLA heard you, it just doesn't care.

Excited to announce our paper, Not All Features Are Created Equal: A Mechanistic Study of Vision-Language-Action Models, was accepted at NeurIPS 2026 🎉
https://t.co/HA1z9UNcvE

[Bryce Grant @bryceagrant1]

@ATTlKA I have a work on mech interp with VLAs and will be releasing some mech interp on WAMs very soon!

[Bryce Grant @bryceagrant1]

@sangoh64166 @sunghopark20124 Hello, good to connect. Interested in interpretsbility for WAMs!

[Bryce Grant @bryceagrant1]

@sunghopark20124 Hey Sungho! Also gonna be at NeurIPS. Working on mech interp for robotic foundation models

[Bryce Grant @bryceagrant1]

We built Action Atlas to catalog VLAs: search 520+ SAEs by concept, steer a feature on live models, and watch hours of rollouts😍.

https://t.co/b0iy608Jga
Code: https://t.co/2zjIl8SBdN

Come visit in Hall 1-4 on Wed, 12/9 at 5pm AEDT in Sydney Poster Session 4!
#NeurIPS2026

[Bryce Grant @bryceagrant1]

Your VLA heard you, it just doesn't care.

Excited to announce our paper, Not All Features Are Created Equal: A Mechanistic Study of Vision-Language-Action Models, was accepted at NeurIPS 2026 🎉
https://t.co/HA1z9UNcvE

[Bryce Grant @bryceagrant1]

Probes decode the instruction at >99%, but the scene outweighs it. With no prompt at all, π0.5 still solves 31.6% of LIBERO-Object. To see what they use instead, our 520+ sparse autoencoders (SAE) split activations into readable features: 79 motion, object, and spatial concepts.

[Bryce Grant @bryceagrant1]

Presenting two mechanistic interpretability works today at @iclr_conf

Giving a talk from 2-2:15 at the Multimodal Intelligence workshop (204C) on our work “Not All Features are Created Equal: A Mechanistic Study of Vision-Language-Action Models.”

Will also be at the Unifying Concept Representation Learning workshop (209) presenting “Gluing Local Contexts into Global Meaning: A Sheaf-Theoretic Decomposition of Transformer Representations.”

Feel free to stop by if you want to talk VLAs, Robotics, or Algebraic Topology!

[Physical Intelligence @physical_int]

To learn more, see more videos, and read a full research paper about Recap and π*0.6, see our blog post here: https://t.co/yZslzlXFbH

[Edward Johns @Ed__Johns]

I'm very excited to finally announce one of the most ambitious projects we've worked on — which makes the front cover of Science Robotics today:

☀️ Learning a Thousand Tasks in a Day ⭐️

Everyday tasks — like those below — can now be learned from a single demonstration each...